what shall we do with the drunken time_t ?

[Bruno Haible]

Hi Paul, all,

I'm trying to port utimens and futimens to native Windows. A major problem
is the semantic of time_t on native Windows. An experiment (at the end of this
mail) shows that:

* While on POSIX systems, time_t is the number of seconds since 1970-01-01
  00:00:00 UTC/GMT [1][2], on native Windows, this is not the case. The
  Microsoft doc is ambiguous [3]
    "The time function returns the number of seconds elapsed since midnight
     (00:00:00), January 1, 1970, Coordinated Universal Time (UTC), according
     to the system clock."
  What the MSVC library returns, is local time minus DST shift.
  In other words, because of
     local_time = utc_time + timezone + dst_shift
  the time_t on native Windows is
     utc_time + timezone

* On mingw, there is additionally a stat() bug. If you have DST today, mingw
  behaves as if you have DST throughout the year.

* The gmtime function is useless: since time_t is (utc_time + timezone),
  it produces a time string that represents (gmt_time + timezone) as well.

What can we do?

  (a) Accept the non-POSIX definition of time_t, and tell programmers not
      to use gmtime() when they want UTC time.

  (b) Make time_t be like on POSIX, by overriding time(), stat(), fstat(),
      and similar functions.

I'm in favour of (b), because
  * Programs that need UTC usually do so because they exchange data with
    other machines, and this messes up with the native Windows notion of time_t.
  * POSIX also has APIs with 'struct timeval' (whose first element is a time_t),
    and utime(), and they also become problematic when time_t is offset by a
    timezone.

Opinions?

Bruno

[1] http://pubs.opengroup.org/onlinepubs/9699919799/functions/time.html
[2] 
http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap04.html#tag_04_16
[3] https://msdn.microsoft.com/en-us/library/1f4c8f33.aspx


====================== Print st_mtime of an existing file ================
#include <sys/types.h>
#include <sys/stat.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>

int main (int argc, char *argv[])
{
        if (argc != 2) { fprintf (stderr, "Usage: stat-file FILE\n"); exit (1); 
}
        const char *filename = argv[1];
        struct stat buf;
        if (stat (filename, &buf) < 0) { perror("stat"); exit(1); }
        time_t tt = buf.st_mtime;
        int day = tt / (24*3600);
        int hour = (tt / 3600) % 24;
        int seconds = tt % 3600;
        fprintf (stdout, "mtime = %d %d %d\n", day, hour, seconds);
        fprintf (stdout, "as GMT: %s\n", asctime (gmtime (&tt)));
        fprintf (stdout, "as localtime: %s\n", asctime (localtime (&tt)));
}
================================ Results ===========================
(I am in CEST, i.e. GMT+1 with DST since end of March.)

A file last touched on 2016-11-27 18:32 GMT:
$ ls -l t.tar
-rw-r--r-- 1 bruno None 10240 Nov 27 19:32 t.tar

Cygwin:
mtime = 17132 18 1920
as GMT: Sun Nov 27 18:32:00 2016         (correct)
as localtime: Sun Nov 27 19:32:00 2016   (correct)

msvc:
mtime = 17132 19 1920
as GMT: Sun Nov 27 19:32:00 2016         (off by 1 h)
as localtime: Sun Nov 27 19:32:00 2016   (correct)

mingw:
mtime = 17132 20 1920
as GMT: Sun Nov 27 20:32:00 2016         (off by 2 h)
as localtime: Sun Nov 27 20:32:00 2016   (off by 1 h)

A file last touched on 2017-04-05 01:49 GMT:
$ ls -l n.txt
-rw-r--r-- 1 bruno None 93 Apr  5 03:49 n.txt

Cygwin:
mtime = 17261 1 2961
as GMT: Wed Apr  5 01:49:21 2017         (correct)
as localtime: Wed Apr  5 03:49:21 2017   (correct)

msvc:
mtime = 17261 2 2961
as GMT: Wed Apr  5 02:49:21 2017         (off by 1 h)
as localtime: Wed Apr  5 03:49:21 2017   (correct)

mingw:
mtime = 17261 2 2961
as GMT: Wed Apr 05 02:49:21 2017         (off by 1 h)
as localtime: Wed Apr 05 03:49:21 2017   (correct)