> You can compute the position (yes, fractional) of the IDFT at
> any single sample...

Im not sure i understand what you mean with "computing the 
fractional position of the IDFT".  IDFTs don't have "positions".  
Windows (ie, window centers or window "corners") may be at a 
certain sample index within a sound file, but that's completely
irrelevant to their definition.  Furthermore, both, the DFT and 
IDFT are (as the "D" in their name indicates) discrete operations 
and thus can't be "positioned" fractionally unless you sample 
rate convert the signal in the first place, which, I believe, 
was the whole point of this discussion.


> just use additive sinusoidal synthesis.

An IDFT takes frequency-domain data and returns a time-domain signal.
Period.  No need for additional additive synthesis. 


> I guess
> you misunderstand what I mean with inverse FT, with a whole IDFT.

I am talking about

           N-1
        1  ---     jtk2pi/N
  x  = --- >    X e
   t    N  ---   k
           t=0

and about the fact that a DFT only not distort the signal iff it is
harmonic (truly and infinitely periodic).  Hence, it is unsuitable
for sample-rate conversion.