#linuxcnc-devel Logs

Jun 04 2018

#linuxcnc-devel Calendar

02:07 AM rmu: skunkworks: i tested some example ngc files thus far, esp. tort.ngc, no problems so far. i suspect exact stop mode could be problematic, because when decelerating to a standstill the last interval ends with velocity and acceleration = 0, and if target position is not reached by then for whatever reason it will hang there forever.
10:23 AM skunkworks: rmu: does it still follow actual program within the G64Px.xxx?
10:25 AM rmu: skunkworks: it should, trajectory should not be changed, i only manipulate the trapezoidal (tangential) speed profile
10:28 AM rmu: skunkworks: i didn't change any of the path blending stuff
10:33 AM skunkworks: huh. (above my pay grade... ) ;)
10:33 AM skunkworks: I can run it through a buch of stuff that I used when we were testing the circular arc blend improvements
10:33 AM * skunkworks is good at breaking things
10:35 AM rmu: skunkworks: have a look at https://forum.linuxcnc.org/media/kunena/attachments/22991/Screenshot_20180531_204051.png vs. https://forum.linuxcnc.org/media/kunena/attachments/22991/Screenshot_20180531_204204.png
10:35 AM skunkworks: I saw that - it looks really cool.
10:35 AM skunkworks: is there any adjustment? (how much it spreads the acc out?)
10:35 AM rmu: not yet
10:36 AM rmu: it spread acceleration so that max accel is nto violated
10:37 AM rmu: "average acceleration" is 8/15 of acceleration is constant case, so i just give only half max accel to the planner
10:37 AM rmu: therefore the planner will not make any plans that can't be executed
10:37 AM skunkworks: ah
10:40 AM skunkworks: git clone https://github.com/rmu75/linuxcnc rmusmooth
10:40 AM skunkworks: heh
10:40 AM rmu: i had some issues with reaching an endpoint exactly with speed 0, so it may well be that there are some problems lurking, especially on a real machine where cycle_time has some jitter
10:41 AM seb_kuzminsky: agreed, this is a topic that has a lot of super fiddly little details with bad failure modes that are hard to get right
10:42 AM seb_kuzminsky: i applaod your fearlessness in tackling it!
10:42 AM skunkworks: well it took a while to get CAB to where it is now - and I think there is still some outstanding bugs :)
10:43 AM seb_kuzminsky: yeah, and i think that's a simpler problem
10:44 AM seb_kuzminsky: so it sounds like microsoft bought github...
10:44 AM seb_kuzminsky: that's... weird
10:45 AM skunkworks: yes - kinda scary
10:45 AM skunkworks: but what are you going to do..
10:45 AM rmu: wait and see
10:45 AM seb_kuzminsky: probably wait and see
10:45 AM seb_kuzminsky: heh yeah
10:45 AM rmu: ;)
10:46 AM skunkworks: sudo make setuid
10:46 AM skunkworks: jeeze
10:46 AM seb_kuzminsky: Password:
10:46 AM skunkworks: too many terminals up
10:46 AM seb_kuzminsky: (let's see if he types it in here)
10:46 AM seb_kuzminsky: in the good old days we hosted our own git server (well, cradek hosted it for us)
10:47 AM rmu: always good to have passwords that are non-suspicious on IRC
10:47 AM seb_kuzminsky: debian.org recently moved all their repos to a self-hosted instance of gitlab
10:47 AM seb_kuzminsky: skunkworks: "too many terminals up" is a great password
10:47 AM skunkworks: heeh
10:49 AM seb_kuzminsky: i like github's issues and code review tools, and i like the "fork" and "star" buttons
10:49 AM seb_kuzminsky: i like how social it is (i bet you thought you'd never hear me say anything like that)
10:50 AM rmu: i don't like the hoops you have to go through to rebase a fork
10:50 AM seb_kuzminsky: there's a great network effect by having so many people/orgs on the same platform, where we can easily see each others work and talk about it and casually collaborate
10:51 AM rmu: for that matter, i didn't figure out how to have a fork of linuxcnc and of machinekit simultanously
10:51 AM seb_kuzminsky: rebase is all local, isn't it? or does github have some kind of rebase button somewhere?
10:51 AM skunkworks: rmu: https://electronicsam.com:2083/cpsess3602608065/frontend/paper_lantern/filemanager/upload-ajax.html?file=&fileop=&dir=%2Fhome%2Felectro%2Fpublic_html%2Fimages%2FKandT%2Ftesting%2Frmu&dirop=&charset=&file_charset=&baseurl=&basedir=
10:51 AM seb_kuzminsky: i have a local git repo with both linuxcnc and machinekit as remotes
10:51 AM skunkworks: heh fail
10:51 AM seb_kuzminsky: though i haven't fetcher machinekit in years
10:51 AM * seb_kuzminsky removes the mk remote from his local repo
10:51 AM skunkworks: http://electronicsam.com/images/KandT/testing/rmu/violation.png
10:52 AM seb_kuzminsky: what are the acc limits on that config?
10:52 AM skunkworks: 30
10:52 AM seb_kuzminsky: ouch
10:52 AM rmu: hmm
10:53 AM skunkworks: I could trigger on one of the violations
10:53 AM seb_kuzminsky: it might be interesting to have a... component or hal circuit or motion debug flag or something that e-stops on violation, to make it easier to debug stuff like that
10:54 AM seb_kuzminsky: then you could run with debugging turned on and see at least what Task was up to when Motion messed up
10:54 AM rmu: interesting
10:55 AM rmu: you didn't change feedrate in flight?
10:55 AM skunkworks: no
10:57 AM rmu: there is something strange going on... :-)
10:57 AM skunkworks: let me run the same program on master just to see
10:57 AM skunkworks: The program is running flat out (maximum feed - only limited by machine constraints)
10:57 AM skunkworks: sudo make setuid
10:58 AM skunkworks: heh
11:02 AM skunkworks: rmu: master http://electronicsam.com/images/KandT/testing/rmu/master.png
11:02 AM skunkworks: is there any config setting?
11:03 AM rmu: no not yet
11:03 AM skunkworks: ok - just making sure
11:07 AM rmu: is this peak values panel available somewhere?
11:08 AM rmu: funny, my halscope also triggers with the "master" planner on excessive y acceleration
11:09 AM seb_kuzminsky: i think it's a custom testing config that skunkworks made, i don't think it ever made it into our repo
11:11 AM rmu: ok, I don't have time right now, but i will investigate what's going on there, that is really strange
11:13 AM skunkworks: rmu: http://electronicsam.com/images/KandT/testing/rmu/violationplit.png
11:14 AM skunkworks: acceleration seems to oscolate
11:16 AM skunkworks: rmu: http://electronicsam.com/images/KandT/testing/rmu/another.png
11:16 AM skunkworks: hmm - for some reason I thought it did..
11:16 AM skunkworks: I thought rob used it.
11:17 AM rmu: that looks like an arc with large radius that blends into one with way smaller radius... seems there is an instability with small velocities ;-)
12:17 PM jepler: re: github -- I also advocate "wait and see" for the project. personally, I can't believe I'll now be using a Microsoft product daily in the course of software development :_/
01:26 PM rmu: hmm. seems i have to have a look at that segment splitting business... that seems to generate these acceleration spikes somehow.
02:00 PM skunkworks: rmu: it is always something isn't it ;)
02:00 PM skunkworks: jepler: I agree. I could be pretty transparent. (or it could go to shit)
02:01 PM * skunkworks looks at rmu's repository for new pushes.. ;)
02:15 PM hazzy-lab: kinda sucks that PyQt signals have to be class attributes instead of instance attributes like Gobject signals..
02:30 PM rmu: so I enabled TC_DEBUG and TP_DEBUG in my build, but i seem to have ordering issues with the printfs
02:30 PM rmu: lines are always complete but it seems some lines go missing and sometimes it is out of order
02:30 PM rmu: how can that be?
02:31 PM rmu: rtapi_print
03:43 PM seb_kuzminsky: are you on rtai or rt-preempt?
04:09 PM rmu: neither nor, non-realtime posix for now
04:09 PM rmu: but that should not make any real difference
04:12 PM rmu: i think i found an issue leading to these acceleration spikes... will have to clean first a little bit to fix this
04:54 PM andypugh: Interesting… https://forum.linuxcnc.org/38-general-linuxcnc-questions/34666-c2-smooth-velocity-profile#111550
04:55 PM andypugh: (May already have been posted here?)
04:55 PM rmu: andypugh: already found some "kinks" ;-)
04:57 PM andypugh: Have you talked to Rob Ellenberg?
05:01 PM rmu: not yet
05:44 PM rmu: so i more or less found where the glitches / acceleration spikes originate, something is not handled correctly when splitting segments
06:03 PM rmu: hmm
08:56 PM skunkworks: rmu: any luck?