Inssues Hole cutting

I can confirm that this is an issue on both QtPlasmaC and plasmac2 in both LinuxCNC 2.9 and master.
I have been able to consistently observe it on metric configs but it does not appear to happen on imperial configs, this is with both configs run both a metric g-code and a imperial g-code.
It does not happen on PlasmaC in LinuxCNC 2.8, nor does it seem to happen on AXIS in LinuxCNC 2.9.
It also happens with all hole detection features commented out, a simple piece of code like below does not run the second G3 line on some holes: Adding a G4 P1 or a G0 X2.873 Y50.627 between the G3 lines appears to make no difference.

Currently I have no idea what is causing it...

In the meantime if you need to cut this job then changing the leadins to perpendicular should work.

 

Thanks Phill for confirming it, if I am going to change to perpendicular, in arc I got better results in the holes but I will have to change to do the job. Thank you anyway.

It's odd, because the code Phill posted, works for me.  But running paulsao's code, I get 3 holes in a row that don't work, so I isolated them and started playing around.  I dont know why, but if I change the starting point coordinates by 0.001mm, it works:

Before:

Result:
 


After:

Result:
 

I do see this issue in the following GUIs:

QtDragon HD XYZ (although not for all the circles)
 

QtDragon XYZ Metric:
 


No idea why, but the following GUIs ran OK:
QtAxis
Axis


Total g-code, because I wonder if these holes work on either of your setups:

Well, I did some digging as I remembered some stuff around arcs and circles and what not.  Long journey short, I came across this:

github.com/LinuxCNC/linuxcnc/commit/ad91...6d2b50d1fff275200941

Andy added the comment:


Which lead me to change line 717 of posemath.h to change the CART_FUZZ value from 1.0e-8 to 1.0e-7, and that worked, the arcs transitioned into the circles as intended.  

Then I noticed that Phill you commented on Andy's commit.  So I changed CART_FUZZ back to 1.0e-8, and made a change at line 1851 of _posemath.c to:

To open the check up a little bit, and that also works.  Since your  commit  and comment on andy's commit had to do with movement around circles, maybe the latter change is more appropriate?

I am honestly not sure which is more appropriate.

Had a play with this, Axis metric sim shows same behavior, goes through arcs only on all 3.
Probably because both G3 lines have the same starting point so nothing to move to, despite different I and J values.
Changing the G0 value changes the result, as it also moves the end point of the first G3 line, so it has where to move to.
Changing the first G3 line X by 0.001 below makes everything work properly, while changing it by 0.001mm above will completely erase the circle. The opposite is true for Y value.
Replacing G3 with G2 in any line works as expected, so does replacing both.

Thanks Tommy.

I do believe this is an issue with the "FUZZ" values not changing with the machine units.

If I import stdio.h into _posemath.c and add a print statement after the line Phill changed in the previous commit (approx line 1851), to print "d", when I run the file, I get the following results...

IMPERIAL CONFIG:



METRIC CONFIG:

Nothing.  The line is never reached because d is not < DOUBLE_FUZZ.

However, if I change the line to compare against CART_FUZZ I get:



The other circle check around line 1862 uses CART_FUZZ as well, so I think changing line 1851 to be the correct route of action...