>However, I still contend that these cases are different.  In the case of a field that goes beyond its normal range, you can clearly
>and unambiguously determine the user's intent, since mktime() was designed to make it easy to do something like calculate a date that
>is, say, 14 days in the future or past by simply adding or subtracting fourteen days on the tm_mday field.   Then mktime() takes the
>denormalized representation and does the right thing.
 
I agree it is unambiguous if this is how a caller uses mktime(). However, a caller could simply set the tm structure and enter an invalid time.  mktime() makes a guess/adjustment in those cases not knowing which direction caller wants to go.
 
Jennifer

From: Scott Atwood [mailto:scott.roy.atwood@gmail.com]
Sent: Thursday, December 11, 2008 11:15 AM
To: Paul Koning
Cc: tz@lecserver.nci.nih.gov; tz@lecserver.nci.nih.gov
Subject: Re: time during standard to DST transition

On Thu, Dec 11, 2008 at 10:59 AM, Paul Koning <Paul_Koning@dell.com> wrote:
>>>>> "Jennifer" == Jennifer Wang <(jennifwa)" <jennifwa@cisco.com>> writes:

 Jennifer> I think in most case when a caller sets the time to 2:10am
 Jennifer> Mar 8, 2009 in America/Los_Angeles, he does not realize
 Jennifer> it's the "missing hour".  If he wants 1:10am (which is a
 Jennifer> valid time), he would use that.  So moving forward to
 Jennifer> 3:10am seems to make sense.

No, it doesn't.  If he meant 3:10 he would have said so.  There is no
basis at all to guess at the user's intent in this way.

It's very simple.  The user is asking for a non-existent time.  So the
reject is valid.  It is every bit as valid as an attempt to set the
current date/time to February 30th.

That's actually a poorly chosen example, since mktime() does guess the users intent in a case like that, by normalizing fields that are out of range.   February 30th would be normalized to March 1st or March 2nd (depending on the year).

However, I still contend that these cases are different.  In the case of a field that goes beyond its normal range, you can clearly and unambiguously determine the user's intent, since mktime() was designed to make it easy to do something like calculate a date that is, say, 14 days in the future or past by simply adding or subtracting fourteen days on the tm_mday field.   Then mktime() takes the denormalized representation and does the right thing.

In the case of the non-existent hour during a DST transition, there is an inherent ambiguity, since you don't know if the caller added to or subtracted from the tm_hour field to get to the invalid time.

-Scott

--
Scott Atwood

Cycle tracks will abound in Utopia.  ~H.G. Wells