#emc-devel | Logs for 2007-02-14

[03:08:57] <SWPadnos_> SWPadnos_ is now known as SWPadnos
[03:15:22] <jepler> I know "dino"'s configuration is a bit strange, but it is true that it gets a following error on the first run of that attached file
[03:15:34] <jepler> I didn't notice anything wrong in the configuration
[03:18:25] <jmkasunich> I suppose I should look
[03:19:06] <jepler> I reduced the problem g-code to this: http://pastebin.ca/354724
[03:19:52] <jepler> increasing the headroom on Z axis more makes the problem go away, but that shouldn't be necessary
[03:20:08] <jmkasunich> 10 meters per minute?
[03:20:12] <jepler> btfoom
[03:20:55] <jmkasunich> his max vel is 7920mm/min, and his f words are asking for 10000
[03:21:05] <jmkasunich> granted, emc should just use the lower of the two rates...
[03:21:10] <jepler> yes it should
[03:24:03] <jmkasunich> grr... won't run
[03:24:10] <jmkasunich> probalby missing something dumb
[03:24:20] <jepler> yeah I had to copy a bunch of stuff around
[03:24:24] <jepler> do you want a tar of my config dir?
[03:24:31] <jmkasunich> sure
[03:24:39] <jepler> (speaking of unhelpful error messages!)
[03:25:07] <jmkasunich> can't load tool table
[03:25:45] <jepler> http://emergent.unpy.net/index.cgi-files/sandbox/mia.tar.gz
[03:27:51] <jmkasunich> thats differnet - I got an unexpected realtime delay message just as it started up
[03:27:57] <jmkasunich> I've _never_ gotten one of those
[03:30:08] <jepler> it has some very aggressive periods specified
[03:30:20] <jepler> I'm running in sim so I don't believe that's the reason for the problem on this person's machine
[03:30:27] <jepler> (sim will just run slow if needed)
[03:30:31] <jmkasunich> I don't either, just commenting
[03:32:13] <jmkasunich> hmm, its running fine now
[03:32:19] <jmkasunich> it ferrored the first time
[03:32:25] <jepler> yes that's what I see too
[03:32:33] <jepler> first run errors, second and following runs are OK
[03:32:37] <jepler> you have to restart to get another error
[03:32:49] <jmkasunich> gotta add some ddt blocks first
[03:32:57] <jepler> http://emergent.unpy.net/index.cgi-files/sandbox/mia-following.png
[03:39:47] <jepler> did you test on 2.1 or on HEAD?
[03:39:55] <jmkasunich> head
[03:40:03] <jepler> I used 2.1
[03:40:39] <jepler> I just had a memory that chris found and fixed a TP bug (requests too much accel) sometime right before or right after the 2.1.0 release
[03:40:56] <jepler> but if you tested TRUNK and I tested 2.1 it's probably not the same bug
[03:41:45] <jmkasunich> I sent the commanded pos from emc thru two ddts to check vel and accel
[03:42:12] <jmkasunich> the limit is 150, the actual is 150 +/- at least 20 in "fuzz"
[03:42:37] <jepler> Issuing EMC_TRAJ_LINEAR_MOVE -- (+220,+88, +0,0.000000,0.000000,-14.600000,0.000000,0.000000,0.000000, +2,132.000000,132.000000,150.000000,)
[03:42:41] <jepler> Issuing EMC_TRAJ_LINEAR_MOVE -- (+220,+88, +0,0.000000,0.000000,-16.100000,0.000000,0.000000,0.000000, +2,132.000000,132.000000,150.000000,)
[03:42:44] <jepler> Issuing EMC_TRAJ_LINEAR_MOVE -- (+220,+88, +0,0.000000,0.000000,-14.600000,0.000000,0.000000,0.000000, +2,132.000000,132.000000,150.000000,)
[03:43:22] <jmkasunich> the significance is?
[03:43:37] <jmkasunich> shouldn't one o fthose numbers be incrementing (move ID)?
[03:43:51] <jepler> the last numbers give the alowable velocity and acceleration computed in userspace
[03:44:03] <jmkasunich> ok, they are exactly what we expected
[03:44:40] <jmkasunich> oh, I just figured out what my noise is on the accel
[03:45:02] <jmkasunich> quantization noise of the float position value, amplified by double differentiation
[03:45:11] <jmkasunich> damn I wish hal floats were actually doubles
[03:45:30] <jepler> I actually expected to see a half-acceleration somewhere in there .. on the reversal, I think
[03:46:07] <jmkasunich> half accel?
[03:46:26] <jepler> yeah
[03:46:57] <jepler> to blend (which is actually not going on here because g61 was specified) the decel phase of one segment is overlapped with the accel phase of the next segment
[03:47:24] <jepler> if the next segment is a reversal, then the accelerations are in the same direction
[03:47:45] <jepler> so unless you give each segment half the possible accel it will sum to way over the full accel
[03:47:50] <jmkasunich> other than a mS or so when the accel changes sign, its always at either +150 or -150
[03:47:52] <jepler> maybe that decision (half vs full) is happening in realtime
[03:49:28] <jepler> does stepgen use floats internally?
[03:49:34] <jmkasunich> hmm, I get pecular screen artifacts when I move the mouse around in the scope screen
[03:49:43] <jepler> yeah I get those too .. someone should figure out what they are :-P
[03:49:45] <jmkasunich> I think it uses a mix of floats and doubles
[03:50:02] <jmkasunich> carefully chosen to work correctly (most of the time)
[03:50:22] <jepler> :-P
[03:50:24] <jmkasunich> some things are worse with double
[03:51:12] <jmkasunich> y'know, this might be a case of the "stepgen goes nutz when accel limit is too low" that we were messing with yesterday
[03:51:30] <jmkasunich> in this case, the limit is only a tiny bit too low, and it only goes a tiny bit nutz
[03:52:04] <jmkasunich> the accel is banging back and forth between + and - limit
[03:52:51] <jepler> does it work for you if the stepgen headroom is increased to 160.0? If so I am going to suggest that to the poster.
[03:53:02] <jmkasunich> I'll try in a minute
[03:53:16] <jepler> OK -- it works for me, but I'd like confirmation before I fire this message off.
[03:53:50] <jmkasunich> ok, I'll give it a shot
[03:55:32] <jmkasunich> got the same error
[03:55:37] <jepler> hmph darn
[03:55:44] <jmkasunich> yeah
[03:55:50] <jmkasunich> and the scope didn't trigger
[03:56:10] <jepler> oh I had put G64 in my nc file, that makes the problem go away too
[03:57:49] <jepler> no it really does fix it for me
[03:57:59] <jmkasunich> 160?
[03:58:02] <jepler> you changed AXIS_2 STEPGEN_MAXACCEL? I guess I just said "headroom"
[03:58:23] <jmkasunich> yes, to 160.0
[03:58:25] <jmkasunich> (just double checked)
[03:58:46] <jmkasunich> the ferrro happens a little later, but it still happens
[03:58:58] <jepler> are you using the poster's file, or my short version?
[03:59:02] <jmkasunich> yours
[03:59:39] <jepler> try a really big value like 200?
[04:01:01] <jmkasunich> worked
[04:01:21] <jmkasunich> (I used 260, lazy typing)
[04:02:16] <jmkasunich> trying 200
[04:02:39] <jmkasunich> that realtime error is 100% consistent, always at startup
[04:03:21] <jmkasunich> 200 works
[04:03:26] <jepler> I don't get the error if I have it go to +Z instead of -Z
[04:03:42] <jepler> er wait I still had my headroom increased
[04:04:07] <jepler> forget it
[04:04:22] <jmkasunich> ferget what?
[04:04:31] <jepler> it happens if the moves are to +Z as well
[04:05:01] <jepler> I was thinking back to the "one step in a surprising direction, particularly at startup" behavior we saw a few months ago
[04:06:35] <jmkasunich> seems to work at 165
[04:06:39] <jmkasunich> fails at 160
[04:06:44] <jmkasunich> 165 = 10% headroom
[04:06:59] <jmkasunich> wish it didn't take so damned long to test
[04:08:55] <jepler> mail sent using 165 as the suggested number
[04:09:00] <jmkasunich> 162 works too
[04:09:12] <jmkasunich> good plan
[04:10:20] <jmkasunich> I noticed something that may just be a coincidence, or may not
[04:10:48] <jepler> uh oh
[04:10:54] <jmkasunich> the velocity ramps negative, then back towards zero as it slows down after the first move
[04:11:07] <jmkasunich> the second move never actually reverses the velocity
[04:11:25] <jmkasunich> even though its an up move
[04:11:33] <jepler> the first two moves are in the same direction, right? the third one is the one that goes up?
[04:11:46] <jmkasunich> duh
[04:11:49] <jmkasunich> I think you're right
[04:12:12] <jepler> I'm done for the night, see you later
[04:12:20] <jepler> don't let this keep you up all night
[04:12:22] <jmkasunich> goodnight
[04:12:24] <jmkasunich> I won't
[04:12:44] <jepler> has the compile farm been calmer lately?
[04:12:55] <jmkasunich> gonna take some pics, make a blog post, and get some sleep
[04:12:59] <jmkasunich> calmer?
[04:13:04] <jepler> less activity?
[04:13:21] <jmkasunich> farm activity is directly proportional to commit activity
[04:13:48] <jmkasunich> which is certainly down since the 2.1 push
[04:13:58] <jepler> I guess I knew that
[04:14:16] <jepler> I was just making conversation really
[04:14:16] <jmkasunich> when all four slots get going, it really bogs down
[04:14:24] <jmkasunich> so the build takes 20-30 mins
[04:14:44] <jmkasunich> and of course, by then somebody's backported the change to the branch, so when that build ends another starts...
[04:15:05] <jmkasunich> the worst is when somebody commits part of a change, then 5 mins later commits the rest
[04:15:23] <jmkasunich> the build starts after the first part, misses the second one, and winds up running twice
[04:15:33] <jepler> yeah
[04:15:45] <jepler> maybe it should wait for "no commits in X seconds"
[04:15:56] <jmkasunich> for instance if somebody fixes one config, then a few mins later realizes that another config has the same problem
[04:16:04] <jepler> there are lots of more sophisticated things, the hard part is guessing which sophisticated thing will make a difference
[04:16:14] <jmkasunich> those are really frustrating since config changes don't really even need a rebuild
[04:16:25] <jepler> "make clean once a day, or after an error"
[04:16:38] <jmkasunich> that could make a huge difference
[04:17:07] <jmkasunich> gotta give that one some serious thought
[04:17:10] <jmkasunich> tomorrow
[04:17:13] <jmkasunich> ;-)
[04:17:14] <jepler> yep
[04:17:19] <jmkasunich> goodnight
[04:17:23] <jepler> see you
[04:31:07] <cradek> hmm looks like I should have been in on that
[04:32:30] <cradek> fwiw, configs/sim/check_constraints.hal is useful for this
[04:47:44] <jmkasunich> don't you just love "wildly wrong" comments?
[04:52:26] <jmkasunich> http://jmkasunich.dyndns.org/shoptask/box-02-13-07.jpg
[04:52:29] <jmkasunich> making some progress
[12:50:24] <jepler> cradek: I would have liked your input too, but you didn't seem to be around
[19:32:36] <SWPadnos> I've been thinking about testing the EMC2 liveCD on a bunch of small computers at ESC
[19:33:04] <SWPadnos> I think the easiest way to do it would be to have a bootable USB memory stick, since most small systems have USB but no CD-ROM
[19:33:28] <SWPadnos> if anyone has any pointers to getting the equivalent of the liveCD onto a memory stick, let me know
[19:33:37] <jepler> I don't know much about bootable USB memory sticks, let me know what you learn :-P
[19:33:38] <skunkworks> bootable usb with emc live would be cool.
[19:34:02] <SWPadnos> note: I specifically want a bootable liveCD-like thing, not an install to a USB key (since I want it to detect video and things on each machine)
[19:34:06] <SWPadnos> jepler, I'll do that :)
[19:34:18] <jepler> maybe you want a USB CDROM drive instead?
[19:34:27] <SWPadnos> that would probably work also
[19:34:39] <SWPadnos> but a USB stick is a lot smaller to carry around
[19:34:48] <skunkworks> Then do it ;)
[19:34:54] <SWPadnos> I'll be doing this at a trade show at the beginning of April
[19:45:50] <skunkworks> SWPadnos: does this give ideas? http://www.damnsmalllinux.org/wiki/index.php/Installing_to_a_USB_Flash_Drive
[19:46:39] <SWPadnos> kind of, but that looks like it's for an install, and I want the liveCD detection features to remain intact. like this: http://www.knoppix.net/wiki/USB_Based_FAQ
[19:47:22] <SWPadnos> I want to run the RT latency tests on a bunch of machines (like PC-104, ITX, mini-ITX, etc), and post the results on the wiki
[19:47:44] <SWPadnos> so I can't do an install, sonce that would set up stuff like X for whatever machine I install on
[19:48:26] <SWPadnos> of course, I may have to bring along a monitor and keyboard (and mouse), so it may be practical to have a USB CD-ROM as well (ir impractical to do the testing :) )
[19:48:32] <SWPadnos> s/ir/or/
[19:48:59] <SWPadnos> oh -the second-to-last section of hte page you linked may be an answer
[19:54:57] <skunkworks> the local university uses a knoppix bootable usb stick to wipe thier computers
[19:55:14] <SWPadnos> the LiveCD has the RT latency testing tools, right?
[19:55:20] <skunkworks> yes
[19:55:48] <SWPadnos> ok. I'm about to tell someone on the Geckodrive list to try it, so they can see why Mach3 is screwing up the charge pump signal :)
[19:56:30] <skunkworks> gecko list is also mach biased isn't it.
[19:56:42] <SWPadnos> fairly, though not as bad as CCED
[19:57:21] <skunkworks> cnczone is also.
[19:57:56] <SWPadnos> yep. it would be nice to have an avertising budget, so sites like that didn't get so biased toward the commercial products
[19:59:21] <skunkworks> we need some millionare back emc2 like the ubuntu guy ;)
[19:59:33] <SWPadnos> heh - maybe someone should call Mark Shuttleworth :)
[20:01:38] <skunkworks> god - he is my age.
[20:01:49] <skunkworks> what the hell have I been doing with my life
[20:01:57] <SWPadnos> heh
[20:02:00] <skunkworks> :)
[20:02:28] <SWPadnos> that means he's significantly younger than me, so what the hell have *I* been doing with my life? :)
[20:02:42] <SWPadnos> I haven't even gone into space
[20:02:55] <skunkworks> he is actually a few months younger.. I had a head start.
[20:04:15] <lerneaen_hydra> how much is he pumping into ubuntu?
[20:04:16] <lerneaen_hydra> gobs?
[20:04:23] <SWPadnos> dunno
[20:04:49] <SWPadnos> I suspect Canonical is close to being profitable, but I don't know what it cost him to keep it running for the last few years
[20:05:38] <SWPadnos> though I have no evidence to support my opinion
[20:06:08] <SWPadnos> time to do some errands aroundthe house. bbiab
[20:32:44] <alindeman> [Global Notice] Hi all .. Services (e.g., ChanServ and NickServ) were down for about 20 minutes. We've put them back online, but unfortunately had to revert to a backup approximately 4 hours old. Hopefully this should not cause many problems, but if it does, please contact staff@freenode.net with your specific issue and we'll review it on a case-by-case basis. Sorry for any inconvenience and thanks for using freenode.
[21:10:07] <christel> [Global Notice] Happy Valentines Day to all! If you're good with web design or graphics and you wish to share some love with freenode and the PDPC, head over to http://freenode.net/news.shtml and have a look.