Day: March 2, 2021

After getting the Sherline mostly working with LinuxCNC 2.8 and the Mesa 5I25 FPGA card, I updated the HAL code turning the Logitech gamepad into The Joggy Thing:

This required significantly more effort than I expected, mainly because I can no longer edit the Eagle schematics defining the HAL file. In the intervening years, Autodesk bought the Eagle EE CAD program, converted it into a subscription service, sutured it onto their Fusion 360 package, and priced the result far beyond my toy budget. While they do offer a free tier limited to “individuals for personal, non-commercial use”, schematics with only two sheets pretty much wipes out its value.

Because the EESchema part of Kicad can export its netlists as XML files, someone experienced in wrangling XSLT, perhaps using Python + lxml, could recreate the function of the Eagle ULP with Kicad schematics / netlists. I am not, however, that person, although it would certainly be a Learning Experience™ of the first water.

So I updated the automatically generated HAL file on hard mode with a text editor, which, given HAL’s limited debugging support, somewhat resembles juggling a greasy bowling ball, a full-throttle chainsaw, and a squalling baby in a poopy diaper.

The major conceptual problem was LinuxCNC’s recent separation of “axes” from “joints”, with resulting changes in both nomenclature and control. Eventually, I found some key hints in a very recent update to a LinuxCNC wiki entry describing a similar Logitech gamepad interface.

The basic Joggy Thing logic remains the same, with “analog” values from the joysticks now presented to both the halui.axis and halui.joint controls. The new trick of holding the pre-startup values (presumably zeros) with feedback around a multiplexer qualifies as a Moby Hack preventing a startup glitch from triggering an error having something to do with an E-stop.

The machine still runs away on X and Z at full throttle instantly after tapping the Machine-On button for the first time in the morning. Come to find out the gamepad starts up with all four joysticks jammed at -1 until the first activation of any axis or button, which I’m sure it always did, but something in HAL’s bowels now responds differently. More work will be required, although I think the simplest solution will involve holding everything inert until the logic sees a specific gamepad button.

The LinuxCNC HAL code as a GitHub Gist: