Interesting discussion, and I don't mean to be an apologist for all things Rigol.
In fact, there is some merit to the original posters complaint and I shouldn't dismiss it.
Firstly, the Rigol calibration square wave is better than I thought and there is almost zero energy beyond the expected odd-harmonics and the fundamental.
I show this in the first image, from my conventional spectrum analyzer. The signal is outside the working bandwidth but still provides a good relative indication of harmonic content.
The perfect square wave has discontinuities whereas the real world square wave does not. However, there is such a thing as a distorted (non perfect) square wave with additional FFT terms.
The second image is from the Agilent 7014B where the user has no control of the sampling parameters. Here, Agilent does a better job of memory management.
Rigol has an Auto memory depth mode that still allows sampling artifacts to leak thru. This would be confusing to the beginner and a source of potential measurement error/confusion.
The FFT can be a gnarly beast to tame. Clearly, there are critical relationships between array size, sample-rate, memory depth, truncations and windowing, etc. where the user has to have a firm
understanding of their sampling system. I am a little surprised that Rigol doesn't shield the user better in Auto mode but I don't find it limiting. I will concede that it can be confusing however.
dan W7NGA