Adapting the documentation of movement limits (feedrate, acceleration, jerk) for delta and other kinematics

[tturpin]

I discovered, by testing, and looking at the code, that the meaning of feedrate, acceleration, and jerk limits for a delta printer (or other non-cartesian kinematics) refers to steppers, and not toolhead position (if I understand correctly). It seems to be known to experienced users and developers, as indicated by MarlinFirmware/Marlin#9403 and MarlinFirmware/Marlin#8778 (the latter already suggested that the documentation should be fixed). However the documentation for M204, M204, and M206 only mentions X Y Z axes, and does not distinguishes between cartesian and other printers. This is missleading, and it would be helpful to clarify it, as it is an important distinction. What about junction deviation (which I didnt' look at) ?

MarlinFirmware/Marlin#8778 fixed the lcd menus, but not the Gcode doc.

Choosing to set limits in terms of steppers or toolhead can be discussed (it seems that Klipper takes the latter approach, or rather a mix of both) and has incidence in terms of implementation effort, but fixing the doc should be simpler (but does require a good understanding of the planner's inner working).

[tturpin]

I would like to add another point. It seems a delta kinematics depends on a non zero, and high enough jerk value to work properly. This seems to be known as trying to disable jerk fails with "CLASSIC_JERK is required for DELTA and SCARA".
I found 3 related issues: #14976, #14978, and #13289. I also observer slow, jerky motion, but in my case, not only with a zero jerk value, but also with values <5mm/s.
It seems that the segments generated to approximate a move by straight moves in the stepper coordinates cannot be sequenced smoothly if the jerk value is too low.

I was thinking that if the segments were too "misaligned", the low jerk could explain the need to decelerate/accelerate between each segment, but if this was the case, I would expect that increasing DELTA_SEGMENTS_PER_SECOND would help, and it doesn't. At the contrary, since the toolhead seems to almost fully stop between each segment, the motion is actually faster with fewer segments per second, since it can reach higher speeds in between, increasing average speed.
Setting DELTA_SEGMENTS_PER_SECOND to a very low value such as 10 makes the stop and go very clear.

For proper configuration of a delta, it would be necessary to understand how this jerk setting is used, how low it can be set without causing this weird behavior. My initial goal was to disable jerk in order to test higher acceleration without mixing the two. but now I'm wondering how my acceleration testing can be meaningful with this other feature that interacts in an unknown way...

About DELTA_SEGMENTS_PER_SECOND, its value is described as a tradeoff, that was probably more challenging before 32-bit boards. Any advice on setting this on newer boards would be nice. Could any noticeable change be expected by increasing my current default value 120 (I've seen 200 elsewhere). If the processor is capable of much more, would it make sense to set it significantly higher to make sure the precision is as high as it can be in this respect ?