Replying to [comment:13 cboos]:
> See attachment:datefmt-r5696.diff
> 
> I've renamed `_normalize` to `to_datetime`, which is reminiscent of `to_unicode` as it serves the same kind of normalization purpose (here: datetime/timestamps -> datetime).

The code looks OK, but the comments still leave me scratching my head.  For example,

{{{
        64	    """Calculate time delta between two `datetime` objects. 
 	65	    (the result is somewhat imprecise, only use for prettyprinting). 
 	66	 
 	67	    If either `time1` or `time2` is None, the current time will be used 
 	68	    instead. 
 	69	
}}}

This code doesn't operate on {{{datetime}}} objects. Like much of the rest of this file, it's a generic function operating on “TimePoints,” where a TimePoint is either:

 a. A {{{datetime}}} object

 b. An {{{int}}}, {{{long}}}, or {{{float}}} in which case it is interpreted ''thus'' (you write).

 c. or {{{None}}}, in which case it is interpreted as “right now.”

OK, we're talking about essentially saying that b) above is unsupported and will cease to work in a future version.  Sorry for the rambling; I lack sleep.  

I guess we still need to decide whether we're supporting c).  Do you really want to allow people to pass {{{None}}} here?  If so, what makes {{{int}}}, {{{long}}}, and {{{float}}} different?  If you don't support that, you can simplify your docs.