I think in most case when a caller sets the time to 2:10am Mar 8, 2009 in America/Los_Angeles, he does not realize it's the "missing hour".  If he wants 1:10am  (which is a valid time), he would use that.  So moving forward to 3:10am seems to make sense.
 
There are also other code such as normalize_overflow() where we make adjustments to what caller wants.
 
Jennifer
From: Scott Atwood [mailto:scott.roy.atwood@gmail.com]
Sent: Wednesday, December 10, 2008 11:26 AM
To: tz@elsie.nci.nih.gov; Jennifer Wang (jennifwa)
Subject: Re: time during standard to DST transition

On Wed, Dec 10, 2008 at 10:43 AM, Jennifer Wang (jennifwa) <jennifwa@cisco.com> wrote:
Hello,
 
We ran into a problem when setting time to the "missing hour" during transition from standard time to DST time.  For example, setting time to Mar 8, 2009 at 2:10am using America/Los_Angeles time zone results in error. 
 
Should mktime() skip forward in this case and set the time to Mar 8, 2009 3:10am?

To me, it seems to make sense to return an error.  The time 2:10am Mar 8, 2009 in America/Los_Angeles really is an ambiguous time.  You can't determine the intension of the caller.  She could either be counting forward one hour from 1:10am, or she could be counting back one hour from 3:10am, which result in different answers.  Likewise, calling mktime() with 2:10am, Nov 2, 2008 in America/Los_Angeles and a negative value of tm_isdst is similarly ambiguous.

I think both of these cases should return an error.

-Scott

--
Scott Atwood

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