[Rd] daylight saving / time zone issues with as.POSIXlt/as.POSIXct (PR#10392)

Prof Brian Ripley ripley at stats.ox.ac.uk
Fri Nov 2 05:58:12 CET 2007 

On Thu, 1 Nov 2007, Peter Dalgaard wrote:

> tplate at acm.org wrote:
>> Running under Windows XP 64 bit, as.POSIXlt()/as.POSIXct() seem
>> to think that US time zones (EST5EDT, MST7MDT) switched from daylight
>> savings back to standard time on Oct 28, 2007, whereas the switch
>> is actually on Sun Nov 04, 2007.
>>
>>
> Not Our Problem. (This sort of thing never is. We are wholly dependent
> on the OS for this information). Check out
>
> http://support.microsoft.com/kb/933360

I think this may well be a long-standing Windows bug that US users have 
not had to suffer before.  From ?as.POSIXlt

      Where OSes describe their valid timezones can be obscure.  The
      help for 'tzset' (or '_tzset' on Windows) can be helpful, but it
      can also be inaccurate.  There is a cumbersome POSIX specification
      (listed under environment variable 'TZ' at <URL:...>
      which is often at least partially supported, but there may be
      other more user-friendly ways to specify timezones. Windows
      documents a specification of the form 'GST-1GDT', but seems always
      to apply the US rules for changing to/from DST with such a
      specification.

Recent versions of Microsoft documentation acknowledge this, e.g.

http://msdn2.microsoft.com/en-us/library/90s5c885(vs.80).aspx

says

  dzn

     Three-letter daylight-saving-time zone such as PDT. If daylight saving
     time is never in effect in the locality, set TZ without a value for
     dzn.  The C run-time library assumes the United States' rules for
     implementing the calculation of daylight saving time (DST).

Now, AFAIK the 'C run-time library' has not been updated to follow US 
politics.


>> Examples:
>>
>> > Sys.timezone()
>> [1] "Mountain Daylight Time"
>> > as.POSIXct("2007-10-30 12:38:47")
>> [1] "2007-10-30 12:38:47 Mountain Daylight Time"
>> > # *** Should report 2007-10-30 14:38:47 EDT:
>> > as.POSIXlt(as.POSIXct("2007-10-30 12:38:47"), "EST5EDT")
>> [1] "2007-10-30 13:38:47 EST"
>> > Sys.time()
>> [1] "2007-11-01 09:22:28 Mountain Daylight Time"
>>
>> > # Bad behavior is manifested in different ways with TZ="MST7MDT"
>> > Sys.setenv(TZ="MST7MDT")
>> > # *** Should report "12:38:47 MDT"
>> > as.POSIXct("2007-10-30 12:38:47")
>> [1] "2007-10-30 12:38:47 MST"
>> > as.POSIXlt(as.POSIXct("2007-10-30 12:38:47"), "EST5EDT")
>> [1] "2007-10-30 14:38:47 EST"
>> > # *** Should report "2007-11-01 09:23:09 MDT"
>> > Sys.time()
>> [1] "2007-11-01 08:23:09 MST"
>> >
>> > sessionInfo()
>> R version 2.6.0 Patched (2007-10-11 r43143)
>> i386-pc-mingw32
>>
>> locale:
>> LC_COLLATE=English_United States.1252;LC_CTYPE=English_United States.1252;LC_MONETARY=English_United States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252
>>
>> attached base packages:
>> [1] stats     graphics  grDevices utils     datasets  methods   base
>> >
>>
>>
>> Furthermore, with the timezone "Mountain Daylight Time"
>> (which is the default I get when I start R), the switch
>> appears to be on Nov 5 in 2006, whereas it actually was
>> on Oct 29 in 2006.
>>
>> > # New R session
>> > Sys.timezone()
>> [1] "Mountain Daylight Time"
>> > # *** wrong switch in 2006 ***
>> > as.POSIXct("2006-10-30 12:38:47")+(-4:7)*(24*3600)
>>   [1] "2006-10-26 12:38:47 Mountain Daylight Time"
>>   [2] "2006-10-27 12:38:47 Mountain Daylight Time"
>>   [3] "2006-10-28 12:38:47 Mountain Daylight Time"
>>   [4] "2006-10-29 12:38:47 Mountain Daylight Time"
>>   [5] "2006-10-30 12:38:47 Mountain Daylight Time"
>>   [6] "2006-10-31 12:38:47 Mountain Daylight Time"
>>   [7] "2006-11-01 12:38:47 Mountain Daylight Time"
>>   [8] "2006-11-02 12:38:47 Mountain Daylight Time"
>>   [9] "2006-11-03 12:38:47 Mountain Daylight Time"
>> [10] "2006-11-04 12:38:47 Mountain Daylight Time"
>> [11] "2006-11-05 11:38:47 Mountain Standard Time"
>> [12] "2006-11-06 11:38:47 Mountain Standard Time"
>> > as.POSIXct("2007-10-30 12:38:47")+(-4:7)*(24*3600)
>>   [1] "2007-10-26 12:38:47 Mountain Daylight Time"
>>   [2] "2007-10-27 12:38:47 Mountain Daylight Time"
>>   [3] "2007-10-28 12:38:47 Mountain Daylight Time"
>>   [4] "2007-10-29 12:38:47 Mountain Daylight Time"
>>   [5] "2007-10-30 12:38:47 Mountain Daylight Time"
>>   [6] "2007-10-31 12:38:47 Mountain Daylight Time"
>>   [7] "2007-11-01 12:38:47 Mountain Daylight Time"
>>   [8] "2007-11-02 12:38:47 Mountain Daylight Time"
>>   [9] "2007-11-03 12:38:47 Mountain Daylight Time"
>> [10] "2007-11-04 11:38:47 Mountain Standard Time"
>> [11] "2007-11-05 11:38:47 Mountain Standard Time"
>> [12] "2007-11-06 11:38:47 Mountain Standard Time"
>> > Sys.setenv(TZ="MST7MDT")
>> > Sys.timezone()
>> [1] "MST"
>> > as.POSIXct("2006-10-30 12:38:47")+(-4:7)*(24*3600)
>>   [1] "2006-10-26 13:38:47 MDT" "2006-10-27 13:38:47 MDT"
>>   [3] "2006-10-28 13:38:47 MDT" "2006-10-29 12:38:47 MST"
>>   [5] "2006-10-30 12:38:47 MST" "2006-10-31 12:38:47 MST"
>>   [7] "2006-11-01 12:38:47 MST" "2006-11-02 12:38:47 MST"
>>   [9] "2006-11-03 12:38:47 MST" "2006-11-04 12:38:47 MST"
>> [11] "2006-11-05 12:38:47 MST" "2006-11-06 12:38:47 MST"
>> > # *** wrong switch in 2007 ***
>> > as.POSIXct("2007-10-30 12:38:47")+(-4:7)*(24*3600)
>>   [1] "2007-10-26 13:38:47 MDT" "2007-10-27 13:38:47 MDT"
>>   [3] "2007-10-28 12:38:47 MST" "2007-10-29 12:38:47 MST"
>>   [5] "2007-10-30 12:38:47 MST" "2007-10-31 12:38:47 MST"
>>   [7] "2007-11-01 12:38:47 MST" "2007-11-02 12:38:47 MST"
>>   [9] "2007-11-03 12:38:47 MST" "2007-11-04 12:38:47 MST"
>> [11] "2007-11-05 12:38:47 MST" "2007-11-06 12:38:47 MST"
>> >
>>
>> I see this behavior on all the Windows systems I have tried:
>> Windows XP 64 bit, Windows XP 32 bit Pro, Windows XP home,
>> Windows 2000, with a variety of R versions.  The systems
>> have all relevant Windows updates applied (unless some were
>> inadvertently missed) and the systems otherwise appear to
>> behave correctly with respect to times and timezones.
>>
>> I do not see this problem on Ubuntu Linux systems.
>>
>> -- Tony Plate
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>
>
>
>

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

More information about the R-devel mailing list