Silliness aside....
If anyone is willing and technically capable. It would be interesting and informative to show the range of errors between a good scope and SG. Various BNC cables, probes with long and short grounding, good/bad terminations, impedance mismatches, etc.
The issue we will run into on the SG/FG will be range.....for these modern scopes an old dinosaur just isn't going to give us 10Hz to 1GHz....and a modern Gen that will do that, and still keep quantization noise and step response phase would cost more than most of the scopes we can afford.
Again seeing perfect square waves isn't that important, in the bigger scheme of things. They are pulses, and really only useful for clocking. We can mangle them pretty bad and still resolve a low jitter clock source.....and if that fails we can buffer and oversample to correct potential dynamic range errors.
I just can't imagine why an utterly pure square wave would be that important. However, where it might become critical, is in low voltage control signal/clock....but again that can be more easily overcome (than measuring the perfect pulse) by simply oversampling and buffering.
I will throw a couple more measurements up, that are somewhat practical. If I have to break out any differential amplifiers and calibrate DC offsets, then it will have exceeded ay kind of real world clock signaling.....a bad PCB layout will utterly destroy rise time, and even a good layout is going to have planar impedance/frequency issues.
Most real world devices have clock signals, that look even worse than the scope plots in this thread....and still manage to work damn well.....even with shit 5mHZ $0.10 crystals we can get very stable 100MHz clocks......this seems to be one place where software and on chip counters has truly trumped the need for pristine analog square waves.
What would interest me is seeing just how low of a signal level we could get to, and still derive a clock.....I think even at 1mV pk-pk, with shit interconnects and ludicrous bandwidth we could still pull a clock from (practically) a sine wave.....
Did you see just how bad the ringing was in the plots I posted earlier? I counted no less than 35 fundamental tones in 1Hz to 100GHz, and that is not even factoring in upper and lower order harmonics. An interesting experiment would be to feed that garbage back out into a clock ref input of a microprocessor or a/d and see how much jitter it really equals.....and then how much we can clean it up with proper buffering/oversampling.....