Configure test for whether localtime caches TZ

Configure test for whether localtime caches TZ

[Ken Brown]

The configure test for whether localtime caches TZ directly modifies the 
environment and so (according to POSIX) might produce unreliable 
results.  Wouldn't it be better to use unsetenv, as in the following 
patch (against the emacs-23 branch):

=== modified file 'configure.in'
--- configure.in        2010-05-10 00:37:59 +0000
+++ configure.in        2010-10-30 18:07:03 +0000
@@ -2441,14 +2441,6 @@
  AC_CACHE_VAL(emacs_cv_localtime_cache,
  [if test x$ac_cv_func_tzset = xyes; then
  AC_TRY_RUN([#include <time.h>
-extern char **environ;
-unset_TZ ()
-{
-  char **from, **to;
-  for (to = from = environ; (*to = *from); from++)
-    if (! (to[0][0] == 'T' && to[0][1] == 'Z' && to[0][2] == '='))
-      to++;
-}
  char TZ_GMT0[] = "TZ=GMT0";
  char TZ_PST8[] = "TZ=PST8";
  main()
@@ -2458,13 +2450,13 @@
    if (putenv (TZ_GMT0) != 0)
      exit (1);
    hour_GMT0 = localtime (&now)->tm_hour;
-  unset_TZ ();
+  unsetenv("TZ");
    hour_unset = localtime (&now)->tm_hour;
    if (putenv (TZ_PST8) != 0)
      exit (1);
    if (localtime (&now)->tm_hour == hour_GMT0)
      exit (1);
-  unset_TZ ();
+  unsetenv("TZ");
    if (localtime (&now)->tm_hour != hour_unset)
      exit (1);
    exit (0);

Re: Configure test for whether localtime caches TZ

[Glenn Morris]

Ken Brown wrote:

> The configure test for whether localtime caches TZ directly modifies
> the environment and so (according to POSIX) might produce unreliable
> results.  Wouldn't it be better to use unsetenv, as in the following
> patch (against the emacs-23 branch):

Seems better, assuming it is portable. Not worth putting in emacs-23
IMO though, since the current code has been there for decades with no
(?) reported problems.

Re: Configure test for whether localtime caches TZ

[Ken Brown]

On 10/31/2010 2:30 PM, Glenn Morris wrote:
> Ken Brown wrote:
>
>> The configure test for whether localtime caches TZ directly modifies
>> the environment and so (according to POSIX) might produce unreliable
>> results.  Wouldn't it be better to use unsetenv, as in the following
>> patch (against the emacs-23 branch):
>
> Seems better, assuming it is portable. Not worth putting in emacs-23
> IMO though, since the current code has been there for decades with no
> (?) reported problems.

It should be more portable than the existing code.  I've checked it into 
the trunk.