Eye Pattern error due to X-axis scale below FLT_EPSILON

[pretoms]

I am trying to generate an eye pattern from a generated WAV file of a 9600-baud signal carrying an AX.25 packet at a 96kHz sample rate.
I feed the baseband waveform into the Clock Recovery (PLL) filter with its symbol rate set to 9600 baud. I connect the same waveform to the Eye Pattern filters din input and connect the recovered clock to its clk input.
The Eye Pattern filter reports this error:
Data rate too low (calculated X axis scale is less then FLT_EPSILON).

The wav file: packets-9600baud-96kHz.wav.zip

Is this a bug or am I using the Eye Pattern in a wrong way?

[azonenberg]

There is currently a lower bound on the data rate the eye pattern filter supports due to loss of precision in temporaries.

If the width of the eye pattern in pixels, divided by the duration of one data bit in femtoseconds, is too small to fit in a float32 without rounding to zero, the eye can't be calculated. We've known about this issue for a long time, the check you're seeing prevents a divide-by-zero crash in this case, but dealing with it has been low priority because most people doing eye patterns are working at much higher data rates where it's not an issue.

I forget the exact lower bound, but it's in the kHz range IIRC.

Keeping this issue open as a reminder that we should work on this eventually.