I was very surprised to find an apparent limitation to a FFT on a simple signal.
Let's consider a timeseries made of L=100 points, and simply composed of the sum of two sinus waves: one with a wavelength of 7 points, the other with a wavelength of 13 points (the phase or amplitude does not matter).
If you compute the FFT of this signal, you won't find exactly the two harmonics, because L is not an exact multiple of 7 or 13.
How do you deal with this problem? Interpolation of the data on a higher resolution?
Edit : this problem can be "minimized" with zero padding