This has been brought up and (rather cavalierly) dismissed before. I challenge anyone to justify the following DateTime behavior:
DateTime('2005-01-01').strftime('%Y-%m-%d') '2004-12-31'
DateTimes are actually timestamps, and the current implementation assumes that a date in ISO8601 format without a timezone specifier is UTC rather than local. It was asserted, last time this came up, that this is compliant with the ISO8601 spec, but my own research shows this to be false. If someone can point me to a truly authoritative source that supports the current behavior, I would appreciate it, but that would not change the fact that the manifested behavior is *broken*. This is a simple two-line fix, folks. I've already done it on my production systems. Cheers, Evan @ 4-am