FWIW, I added a Mesa output to my HAL, connected to the same signal used for motion.probe-input in order to monitor it with the oscilloscope along with the touch probe's trigger signal.
The total delay is 2-3ms (see attachment;CH1 is the probe trigger signal, CH2 is the Mesa output).
The reason it's not ~1ms (servo period) is that I had an OR gate connected to motion.probe-input for both a touch probe and a tool setter and the OR's function was executed after hm2_7i92.0.write. Removing the OR gate and connecting the Mesa input used for the touch probe directly to motion.probe-input (and the new output) dropped the delay to 1-2ms (I don't have a screenshot for that).
Either way, even 3ms delay does not explain the pretravel variation I'm observing, so the problem must be somewhere else. At 50mm/min probing speed, the total error introduced by that delay should be around 2.5um (given that I didn't make a mistake in my calculations). So not an actual problem at the moment.
Looking at my HAL file (initially generated by PNCconf) I wondered whether I need the PIDs (EDIT: I'm using open-loop steppers). After searching the forums I found that it helps with position estimation (if I understood correctly). A couple of questions:
1. How do the PIDs affect the position recorded during probing?
2. Is the following snippet correct or it's a bug in PNCconf?
net x-output <= pid.x.output
net x-output <= hm2_7i92.0.stepgen.00.velocity-cmd
Doesn't this mean that x-output has 2 writers?
