Misc #10473: Date.to_datetime.to_time != Date.to_time

• Status: Open
• Priority: Normal
• Assignee: -

[lojack (Jack Lowe)]

Date.new(2014,1,1).to_datetime.to_time.utc.to_s
=> "2014-01-01 00:00:00 UTC"

Date.new(2014,1,1).to_time.utc.to_s
=> "2014-01-01 05:00:00 UTC"

[yui-knk (Kaneko Yuichiro)]

Date#to_time interprets date as the local time zone.
But Date#to_datetime interprets date as UTC.

I think this incompatibility is a king of bug.
We have two ways to solve this incompatibility, to make both methods to interpret date as the local time zone,
or to interpret date UTC.

It is difficult to judge which is more natural or useful.
But I always interpret date as the local time zone in my daily life...

diff --git a/ext/date/date_core.c b/ext/date/date_core.c
index 3a10fcb..4cc3dc9 100644
--- a/ext/date/date_core.c
+++ b/ext/date/date_core.c
@@ -8520,35 +8520,7 @@ date_to_date(VALUE self)
 static VALUE
 date_to_datetime(VALUE self)
 {
-    get_d1a(self);
-
-    if (simple_dat_p(adat)) {
-       VALUE new = d_lite_s_alloc_simple(cDateTime);
-       {
-           get_d1b(new);
-           bdat->s = adat->s;
-           return new;
-       }
-    }
-    else {
-       VALUE new = d_lite_s_alloc_complex(cDateTime);
-       {
-           get_d1b(new);
-           bdat->c = adat->c;
-           bdat->c.df = 0;
-           RB_OBJ_WRITE(new, &bdat->c.sf, INT2FIX(0));
-#ifndef USE_PACK
-           bdat->c.hour = 0;
-           bdat->c.min = 0;
-           bdat->c.sec = 0;
-#else
-           bdat->c.pc = PACK5(EX_MON(adat->c.pc), EX_MDAY(adat->c.pc),
-                              0, 0, 0);
-           bdat->c.flags |= HAVE_DF | HAVE_TIME;
-#endif
-           return new;
-       }
-    }
+    return time_to_datetime(date_to_time(self));
 }

 /*
diff --git a/test/date/test_date_conv.rb b/test/date/test_date_conv.rb
index 3729476..0feaf2a 100644
--- a/test/date/test_date_conv.rb
+++ b/test/date/test_date_conv.rb
@@ -126,10 +126,13 @@ def test_to_datetime__from_time

   def test_to_datetime__from_date
     d = Date.new(2004, 9, 19) + 1.to_r/2
-    d2 = d.to_datetime
-    assert_equal([2004, 9, 19, 0, 0, 0, 0, 0],
-                [d2.year, d2.mon, d2.mday, d2.hour, d2.min, d2.sec,
-                 d2.sec_fraction, d2.offset])
+
+    with_tz('Asia/Tokyo') do
+      d2 = d.to_datetime
+      assert_equal([2004, 9, 19, 0, 0, 0, 0, (3.to_r/8)],
+       [d2.year, d2.mon, d2.mday, d2.hour, d2.min, d2.sec,
+        d2.sec_fraction, d2.offset])
+    end
   end

   def test_to_datetime__from_datetime

[akr (Akira Tanaka)]

The proposed patch seems fine.

However I recommend to add more tests for old dates around transition between Jurian to Geregorian Calendar.

[akr (Akira Tanaka)]

Akira Tanaka wrote:

The proposed patch seems fine.

However I recommend to add more tests for old dates around transition between Jurian to Geregorian Calendar.

I found that there are days that exists on Jurian carendar but not on Gregorian calendar.

1000/2/29 is exist on Jurian calender but it is not exist on Gregorian calendar.

So, Date.new(1000, 2, 29) preveserves the arguments but Time.new(1000, 2, 29) doesn't.

% ruby -rdate -e '
p Date.new(1000, 2, 29) 
p Time.new(1000, 2, 29)
'
#<Date: 1000-02-29 ((2086367j,0s,0n),+0s,2299161j)>
1000-03-01 00:00:00 +0918

So, Date.new(1000, 2, 29).to_time.to_datetime doesn't preserve the arguments
(and Date.new(1000, 2, 29).to_time.to_datetime.to_date doesn't round trip).

% ruby -rdate -e '
d = Date.new(1000, 2, 29)
p d
p d.to_time
p d.to_time.to_datetime  
p d.to_time.to_datetime.to_date
'
#<Date: 1000-02-29 ((2086367j,0s,0n),+0s,2299161j)>
1000-03-01 00:00:00 +0918
#<DateTime: 1000-03-01T00:00:00+09:18 ((2086367j,52861s,0n),+33539s,2299161j)>
#<Date: 1000-03-01 ((2086368j,0s,0n),+0s,2299161j)>

[akr (Akira Tanaka)]

Similar problem exists on Samoa (Pacific/Apia).

There is no 2011-12-30 in Pacific/Apia.
http://en.wikipedia.org/wiki/International_Date_Line

So, Date.new(2011,12,30) preserves the arguments but Time.new(2011,12,30) doesn't.

% TZ=Pacific/Apia ruby -rdate -e 'p Date.new(2011,12,30), Time.new(2011,12,30)'
#<Date: 2011-12-30 ((2455926j,0s,0n),+0s,2299161j)>
2011-12-31 00:00:00 +1400

Date doesn't depend on Time as much as possible.
In this sense, the current behavior, Date#to_datetime chooses UTC, is reasonable.

But if it is too confusing and Date#to_datetime should respect the local time zone
using Time, it is better to use only utc_offset as follows instead of the cascading
conversion date.to_time.to_datetime.

% TZ=Pacific/Apia ruby -rdate -e '
class Date
  def to_datetime2
    DateTime.new(year, mon, mday, 0, 0, 0, Time.new(year, mon, mday).utc_offset/86400r, start)
  end
end
d = Date.new(2011,12,30)
p d.to_datetime2'
#<DateTime: 2011-12-30T00:00:00+14:00 ((2455925j,36000s,0n),+50400s,2299161j)>