Back
[07:17:18] <skunkworks> seb_kuzminsky, did you email rob about the difference between 2.7 and his branch? My emails stopped with rob not understanding the difference.
[09:12:58] <skunkworks> pcw_home, is there a how-to for installing rt-preempt on precise?
[09:13:24] <pcw_home> I think its just a package
[09:13:45] <skunkworks> Hmm - on wheezy it is in the package manager - I could not find it in precise..
[09:14:14] <cradek> on wheezy it's spelled linux-image-rt-686-pae or linux-image-rt-amd64
[09:14:34] <skunkworks> I will look around
[09:14:43] <pcw_home> yeah, well on Trusty I had to make my own kernel
[09:15:25] <pcw_home> (which is trivial compared to RTAI futzing)
[09:16:41] <skunkworks> all I see is something called lowlatency
[09:17:25] <skunkworks> linux-image-lowlatency-pae
[09:17:29] <cradek> you could probably install the debian kernel
[09:17:52] <skunkworks> or I could just install wheezy..
[09:18:08] <cradek> well sure
[09:18:21] <cradek> but only if you want to make your life better
[09:19:01] <skunkworks> heh
[09:22:38] <seb_kuzminsky> there is no rt-preempt kernel in ubuntu
[09:22:56] <seb_kuzminsky> they don't ship one, i mean, you have to roll your own
[09:23:56] <skunkworks> seb_kuzminsky, did you get my message above?
[09:24:45] <seb_kuzminsky> yeah
[09:25:11] <seb_kuzminsky> last night i started documenting what i meant
[09:25:19] <seb_kuzminsky> here's what 2.7 does:
http://highlab.com/~seb/linuxcnc/2.7-radius-tolerance.png
[09:25:51] <skunkworks> ooh...
[09:26:38] <skunkworks> graphs..
[09:27:49] <seb_kuzminsky> so, each arc is specified with two radii, start & finish
[09:28:01] <seb_kuzminsky> one is bigger than the other, and it doesnt matter which is which
[09:28:10] <seb_kuzminsky> the big radius is on the x axis, small on y
[09:29:02] <seb_kuzminsky> for a given big radius on x, the valid values for the small radius must be both:
[09:29:14] <seb_kuzminsky> 1. above the red line, and
[09:29:27] <seb_kuzminsky> 2. above the lower of the green & blue lines
[09:31:01] <seb_kuzminsky> blue & green cross at r_big = 0.5
[09:31:22] <seb_kuzminsky> red & green cross at r_big = 50
[09:31:42] <seb_kuzminsky> that defines the set of arcs accepted by 2.7
[09:32:02] <seb_kuzminsky> i need to do the same for rob's branch, maybe tonight
[09:33:05] <seb_kuzminsky> he does away with one of those three lines, which of course changes the set of arcs accepted
[09:33:23] <seb_kuzminsky> *that can be accepted, by tuning the tolerance & relative-error settings
[09:33:29] <seb_kuzminsky> bbl
[11:38:46] <seb_kuzminsky> http://yorik.uncreated.net/guestblog.php?2015=22
[11:38:49] <seb_kuzminsky> hi rob_ellenberg
[11:39:02] <rob_ellenberg> hi seb, how's it going?
[11:39:19] <seb_kuzminsky> i'm working on a document describing my concern about the change in radius tolerance, but it's not ready yet
[11:40:19] <seb_kuzminsky> i have these plots, but i'm working on some words to go with them:
[11:40:21] <seb_kuzminsky> http://highlab.com/~seb/linuxcnc/2.7-radius-tolerance.png
[11:40:26] <seb_kuzminsky> http://highlab.com/~seb/linuxcnc/rob-radius-tolerance.png
[11:41:14] <seb_kuzminsky> in 2.7 there are three regions, with three different requirements
[11:41:21] <seb_kuzminsky> r_big < 0.5: r_small > r_big - 0.0005
[11:41:25] <seb_kuzminsky> r_big < 50: r_small > 0.999 r_big
[11:41:31] <seb_kuzminsky> r_big > 50: r_small > r_big - 0.05
[11:41:42] <seb_kuzminsky> in the spiral-arc branch there are two regions:
[11:41:47] <seb_kuzminsky> r_big < 0.5: r_small > 0.999 r_big
[11:41:52] <seb_kuzminsky> r_big > 0.5: r_small > r_big - 0.0005
[11:42:37] <seb_kuzminsky> i'm not saying the way 2.7 does it is perfect, i'm just pointing out that the set of accepted arcs is changing, and some existing gcode programs won't work any more
[11:43:00] <seb_kuzminsky> maybe it's for the better, but we should know about it and talk about it and be prepared to explain/defend the change to users
[11:43:15] <rob_ellenberg> That's odd, the whole point of this branch is to increase the allowed tolerances
[11:43:29] <rob_ellenberg> maybe I screwed up the particular tolerance numbers?
[11:43:48] <seb_kuzminsky> for the analysis there i set the tolerance numbers to the same values they had in 2.7
[11:43:57] <seb_kuzminsky> maybe i screwed up my analysis...
[11:44:05] <rob_ellenberg> Ahh, now I see
[11:44:13] <seb_kuzminsky> ?
[11:44:32] <rob_ellenberg> The differences make more sense if you set the values to be the same
[11:45:14] <seb_kuzminsky> the gcode language we try to implement does not allow spiral arcs, so while it's good that motion deals correctly with them, the interpreter should still reject them (because they indicate an invalid program)
[11:45:35] <seb_kuzminsky> i wish i could stay & talk about this, but i need to pay attention to my dayjob
[11:45:47] <rob_ellenberg> ok, I'll ttyl
[11:45:48] <seb_kuzminsky> i'll read back later, and i'll be online around 9 or 10 tonight (mountain time)
[11:45:58] <rob_ellenberg> Ok, I'll be on then
[11:46:30] <seb_kuzminsky> here's a partially formatted document with my thinking in it:
http://highlab.com/~seb/linuxcnc/radius-tolerance.html
[11:48:12] <rob_ellenberg> in your calculations, try setting spiral_abs_tolerance to 100 * 0.0005
[11:48:18] <rob_ellenberg> that's what it's meant to represent
[11:49:29] <rob_ellenberg> wait, nm, I see what you did
[16:15:08] <mozmck> pcw_home: you around?
[16:20:19] <skunkworks> pcw_home: the laptop I posted - I just installed wheezy and it runs the rtai kernel fine <30k but the rt-preempt kernel doesn't play well. >500us
[16:20:33] <skunkworks> I have not played with any of the bios settings
[16:56:44] <seb_kuzminsky> finished cleaning up my radius tolerance notes:
http://highlab.com/~seb/linuxcnc/radius-tolerance.html
[17:13:38] <seb_kuzminsky> it's my opinion that the branch should not be merged as is
[17:14:16] <seb_kuzminsky> i would welcome a branch with the (unrelated) tp fixes from that branch
[17:14:57] <seb_kuzminsky> and another branch that has the spiral-arc improvement but restores the current small-arc radius tolerance behavior
[17:20:52] <PCW> mozmc: yeah
[17:21:59] <PCW> skunkworks that a bit disappointing I will get a chance to play with a Dell second gen I5 laptop in a week or so
[17:24:26] <PCW> On the other hand I've got ~3 weeks 24/7 at 4 Khz on a H81-D3 no issues
[18:48:56] <KGB-linuxcnc> 03Sebastian Kuzminsky 052.7 b6a4822 06linuxcnc 03tests/motion/jogwheel/.gitignore tests: ignore parameter save files * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=b6a4822
[18:51:36] <KGB-linuxcnc> 03Robert W. Ellenberg 052.7 eecbfa2 06linuxcnc 10src/emc/tp/tp.c tp: fix for incorrect estimation of final velocity when deciding ramped vs. trapezoidal acceleration * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=eecbfa2
