#linuxcnc-devel | Logs for 2014-07-23

[02:49:02] <seb_kuzminsky> linuxcnc-build: back so soon?
[04:09:52] <seb_kuzminsky> jepler: i have the odroid u3 you linked earlier, it's been reliable for me and reasonably fast
[04:10:05] <seb_kuzminsky> had it in the buildbot for a while, sounds like it's time to add it back in again
[04:25:55] <seb_kuzminsky> ./configure --with-realtime=uspace --disable-build-documentation; time make -j6
[04:26:07] <seb_kuzminsky> 4 minutes, 45 seconds
[04:35:39] <seb_kuzminsky> Runtest: 142 tests run, 142 successful, 0 failed + 0 expected
[07:08:06] <jepler> seb_kuzminsky: oh hey, that's good to know.
[07:08:19] <jepler> beaglebone black running from sdhc: real 46m42.460s
[07:12:27] <jepler> milltask takes 25% CPU on bbb
[07:13:00] <jepler> I forget if that something I came up with a patch for when I tinkered with ubc3 last winter, or if I just had some improvements for axis...
[07:22:59] <jepler> Runtest: 142 tests run, 142 successful, 0 failed + 0 expected
[07:29:14] <CaptHindsight> there's also crosstool ng to speed up builds
[07:33:48] <jepler> Is the performance difference between the slow ARMs I have (AM355x on BBB and Zynq-7000 on Parallela) and the fast ARMs seb and I have (Exynos) really 8-10x?
[07:34:16] <jepler> or is it something about the system beyond the CPU, like RAM?
[07:34:52] <jepler> specifically, performance at building software...
[07:36:36] <CaptHindsight> probably a combination of both, multiple cores with 2-4x the core speed + 1-2GB ram vs 512
[07:37:16] <jepler> the chromebook also is much lower CPU load running linuxcnc
[07:37:29] <jepler> milltask <15%, rtapi_app <10%, axis <5%
[07:37:38] <jepler> vs 20+%, 20+%, 50+% on bbb
[07:38:34] <jepler> I was wondering more about RAM bandwidth and latency than capacity
[07:39:07] <jepler> caches too, gcc speed was known to be heavily dependent on CPU cache size several years ago
[07:39:13] <CaptHindsight> well your chromebook is 1.7GHz dual core A15
[07:40:26] <CaptHindsight> RAM bandwidth as well, I'll have to look back at the docs if you want the numbers
[07:40:46] <jepler> oh I don't need specific numbers
[07:41:23] <CaptHindsight> exynos 32KB (Instruction)/32KB (Data) L1 Cache and 1MB L2 Cache 2-ports 32-bit 800MHz LPDDR3/DDR3 Interfaces
[07:41:41] <jepler> but if you've got 'em :)
[07:44:14] <CaptHindsight> AM335x 32KB of L1 Instruction and 32KB of Data Cache 256KB of L2 Cache 16-bit (LPDDR-400, DDR2-532, DDR3-800)
[07:45:16] <CaptHindsight> so 32b vs 16b RAM interface and much larger L2 cache on the Exynos
[07:46:48] <CaptHindsight> and if the chromebook used both DDR3 ports then it's 4X the bandwidth over the BBB
[07:49:38] <CaptHindsight> 1 GHz single core A8 on the BBB, 1.7GHz dual core A15 on the Exynos
[07:49:48] <jepler> the board shot I have for the chromebook has 8 chips marked elpida j2108edbg
[07:50:10] <jepler> they're x8 chips, so it must use both channels
[07:50:48] <jepler> bbl, need to go to $DAY_JOB
[08:19:00] <skunkworks> logger[psha],
[08:19:00] <logger[psha]> skunkworks: Log stored at http://psha.org.ru/irc/%23linuxcnc-devel/2014-07-23.html
[08:19:34] <KGB-linuxcnc> 03Jeff Epler 05master f32900c 06linuxcnc 10src/configure.in configure: Don't suggest --enable-simulator * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=f32900c
[08:28:09] <mozmck> I had to install python-gst0.10 to get gmoccapy to run, but did not see that as a run dependency in control.in
[08:30:47] <jepler> I don't know that norbert reads IRC
[08:30:53] <jepler> I don't know how best to communicate with him
[08:31:19] <mozmck> ok, maybe I'll email him - I didn't know the best place to mention it.
[08:31:27] <jepler> I agree, it appears gmoccapy and gscreen at least optionally require gst
[08:31:58] <jepler> or you can just fix control.in yourself
[08:33:01] <mozmck> ok, I'm not sure it the package name will be the same on other systems? I'm using linuxmint 16
[08:33:11] <jepler> I checked the package name on ubuntu 10.04 and debian 7
[08:33:13] <jepler> it's the same
[08:33:17] <mozmck> ok
[08:33:19] <mozmck> thanks
[08:37:45] <mozmck> do I need to commit the change first and then use git format-patch?
[08:38:10] <cradek> yes
[08:38:16] <cradek> after committing, git format-patch -1
[08:38:23] <cradek> will do it
[08:44:54] <mozmck> Should I send it to the dev list?
[08:45:25] <cradek> nah, one of us can just push it for you
[08:45:36] <cradek> I mean, I can
[08:45:48] <mozmck> ok, I'll email it to you.
[08:45:58] <cradek> ok, cool
[08:46:18] <mozmck> seems like I had push access a long time ago, but I'm sure my ssh keys are different now and all.
[08:46:29] <mozmck> not like I've done much either :)
[09:12:10] <mozmck> in master, is robs planner in tp.c? is the old planner still in there as well?
[09:14:24] <CaptHindsight> how do we test the floating point bug on ARM?
[09:14:44] <CaptHindsight> to best see how it effects motion
[09:16:40] <jepler> CaptHindsight: it's hard to trigger deliberately, from what I can tell
[09:18:12] <cradek> mozmck: I can update your key for you anytime
[09:18:19] <mozmck> heh, you probably have to be several hours into a multi-thousand dollar piece of metal before it will trigger and destroy the work :)
[09:18:29] <jepler> at one point, I wrote a (userspace, pthreads) program which readily showed the problem on i386 linux when not using -Os to ensure it used "fst(p)l" to store doubles http://emergent.unpythonic.net/01189609097
[09:18:33] <jepler> but I can't find the program itself
[09:18:43] <jepler> and I never actually demonstrated a problem while running linuxcnc
[09:22:20] <KGB-linuxcnc> 03Moses McKnight 052.6 c58fad6 06linuxcnc 10debian/control.in New runtime dependency for gmoccapy * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=c58fad6
[09:23:50] <cradek> mozmck: thanks for sending that
[09:24:01] <mozmck> no problem.
[09:24:25] <mozmck> did you see my PM?
[09:24:58] <seb_kuzminsky> jepler: running sim/axis from master on my u3, with Axis displaying over X on my crappy home wifi, i get axis taking 20-25% CPU, sshd 10-15%, and milltask 3-5%
[09:25:03] <seb_kuzminsky> that's while running the splash code
[09:25:27] <skunkworks> mozmck, my understaning is that his new tp touches more than just tp.c.. It is a pretty big restructure.
[09:27:44] <mozmck> skunkworks: ok, I was just curious to look at the code and not sure where it is. I see the new blendmath and spherical_arc files.
[09:28:53] <skunkworks> The previous behavior is there (parabolic blending) and you can turn off circular arc blending.. but I think even that was restructured.
[09:29:18] <skunkworks> (and a few slight acceleration violations where fixed)
[09:29:55] <mozmck> ok. Probably don't have time to look at it much now anyhow, thanks.
[09:29:59] <cradek> mozmck: it's true that all the old code/algorithm remains, because the new planner falls back to the old code for the weird/hard cases
[09:33:54] <skunkworks> mozmck, rob gave a talk on the new TP
[09:33:56] <skunkworks> http://www.youtube.com/watch?v=412N5A-N8Fc
[09:34:15] <jepler> CaptHindsight: http://emergent.unpythonic.net/files/sandbox/atomy.cc illustrates the problem on an intel desktop, but to my surprise not on my chromebook.
[09:43:48] <CaptHindsight> jepler: where in linuxcnc did you mentioned the instructions are used/ was it just the trajectory planner ?
[09:48:20] <jepler> CaptHindsight: anytime a value is stored or written via a hal_float_t (which is a typedef for volatile double)
[09:54:03] <jepler> boy, I need some time to start to grok arm assembly
[09:54:30] <jepler> to someone who is familiar with x86 and knew mips back in the day, this is just moon language: ldmibls r9, {r1, r3, r4, r7, r8, r11, r12, pc}
[09:54:45] <jepler> I am guessing it restores a bunch of registers
[09:57:15] <pcw_home> load mibbles?
[09:57:42] <ssi> I used to do arm assembly many years ago
[09:58:12] <ssi> the arm set may be vastly different now though :/
[09:58:13] <jepler> load multiple "IB" from the address in r9. not sure what IB is
[09:58:54] <ssi> I think that's increment before
[09:59:01] <ssi> vs increment after, would be the ldmia instructions
[09:59:32] <jepler> ssi: ah, the stack can flow two ways and can be pre- or post-incremented
[09:59:38] <ssi> so load multiple, increment before, branch if less or same
[09:59:38] <ssi> yea
[09:59:45] <jepler> yes, the cheatsheet I have lists IA/IB/DA/DB but without explanation
[10:00:29] <jepler> maybe that instruction is returning from a function, then, since it is popping the program counter
[10:00:46] <jepler> but r9 is not sp, so maybe not
[10:01:15] <pcw_home> so it has a bitmask of registers to load?
[10:01:31] <jepler> pcw_home: seems like that must be how it's encoded
[10:01:32] <ssi> so I think what should happen is, r1 will get the value of r9 + 4, r3 will get r9 + 8, etch
[10:03:32] <ssi> if you were loading from stack, you'd use ldmfa/fd/ea/ed instead
[10:03:45] <ssi> ldmib loads from the pointer r9 in this case
[10:03:53] <jepler> r9's not otherwise read or written by this whole function
[10:04:16] <jepler> is it a special register?
[10:04:31] <jepler> "the meaning of this register is defined by the platform standard"
[10:04:35] <ssi> haha
[10:05:11] <jepler> "it may designate it as the thread register (TR) in an environment with thread-local storage"
[10:05:15] <ssi> >From the ABI documentation I've read, register r9 is in some way
[10:05:16] <ssi> reserved for implementation of things like thread local storage.
[10:05:49] <ssi> r9 might be a callee-save register or not (on some variants of AAPCS it is a special register)
[10:05:57] <jepler> CaptHindsight: fwiw I am not surprised atomy.cc doesn't readily trigger on any non-SMP ARM
[10:07:10] <jepler> because the context switch has to happen on exactly the right instruction
[10:07:32] <jepler> or, in the case of ldrd on arm, exactly in the middle of the right instruction
[10:08:42] <jepler> is it possible that the exynos has a stronger atomicity guarantee than the regular arm platform? are the docs which would say so public?
[10:09:00] <CaptHindsight> let me check samsung
[10:12:44] <jepler> here's another interesting wrinkle: in my test program on chromebook, "shared" ended up 16-byte aligned
[10:13:05] <jepler> if I cause it to be 8-byte-aligned or less, it gets a read of a mixed value readily
[10:17:37] <jepler> .. if 'shared' is at address 0x10918 or 0x10914 the problem is detected; if it's at 0x10910 the problem's not detected
[10:17:50] <jepler> I assume the platform alignment for doubles is either 4 or 8
[10:21:49] <CaptHindsight> Exynos 5 Dual 5250 is ARM Coretx A-15 only, vs also big-little and A7 like the quad and octa cores
[10:30:31] <CaptHindsight> http://www.samsung.com/global/business/semiconductor/file/product/Exynos_5_Dual_User_Manaul_Public_REV100-0.pdf the only Public doc I can find for now, it's mostly on hardware registers
[11:07:38] <seb_kuzminsky> linux 3.10 got big.LITTLE "heterogeneous multi-processing" scheduler support
[11:12:04] <jepler> it's so stupid about ARM that we peons have to resort to bootleg copies of the ARM Architecture Reference Manual (why is that not abbreviated ARM-ARM?)
[11:12:42] <cradek> seb_kuzminsky: I made the same changes in iocontrolv2. I've got that fix and removal of stray execute bits ready, and I don't remember if you said yes for 2.6
[11:15:38] <cradek> http://timeguy.com/gitweb?p=linuxcnc.git;a=log;h=refs/heads/proposed
[11:17:52] <seb_kuzminsky> cradek: i didn't look closely at your perms fix, but i like having correct perms, so pls push that to 2.6
[11:18:08] <seb_kuzminsky> i'll look at the io2 commit
[11:18:55] <seb_kuzminsky> jepler: yeah that's really crazy, it's like they don't want us (well, you) to write software for their hardware
[11:19:00] <cradek> I don't know whether people are using iov2
[11:19:12] <seb_kuzminsky> i think there's some sample config that uses it
[11:19:19] <cradek> but it once I got over my revulsion, it was easy to make the same change
[11:19:23] <cradek> s/it//
[11:21:06] <seb_kuzminsky> sim/axis/orphans...
[11:21:09] <seb_kuzminsky> and sim/axis/iocontrolv2
[11:22:20] <jepler> anyway, I'm going to put aside this ARM atomic double-precision thing again. The status quo, in which there are ALWAYS at least 21 correct bits of mantissa and ALMOST ALWAYS 53 correct bits of mantissa is going to be good enough in practice
[11:22:38] <jepler> compared to ALWAYS 23 correct bits if we went back to hal_float_t being a 32-bit float instead of a 64-bit double
[11:23:54] <jepler> .. though based on my test today, it might be an architectural feature of some CPUs that 16-byte aligned 8-bit stores via STRD are "more atomic" than others; if so, we could arrange for hal to 16-byte align doubles.
[11:24:01] <jepler> at least, doubles that are signals
[11:31:49] <seb_kuzminsky> seems totally reasonable
[11:32:54] <CaptHindsight> https://silver.arm.com/download/download.tm?pv=1603196 ARM Architecture Reference Manual ARMv7-A and ARMv7-R edition
[11:33:01] <CaptHindsight> registration required
[11:37:04] <jepler> er, 8-byte stores, of course
[12:43:17] <memfrob> http://www.xenomai.org/index.php/Xenomai:Roadmap#Toward_Xenomai_3 that's awesome!
[13:31:59] <jepler> memfrob: interesting
[13:34:47] <PCW> usb/ether ran for a couple hours at 2 KHz then got a RT error of about 8 ms :-(
[13:35:12] <skunkworks> PCW, you sound suprised... ;)
[13:35:59] <PCW> I was surprised it worked at all above 100 Hz
[13:36:39] <PCW> well maybe 100Hz is about right with random 8 ms delays
[13:38:17] <CaptHindsight> lets turn the tables, make RTAI run as SMI! :p
[13:40:12] <memfrob> jepler, i thought so too :)
[13:43:02] <PCW> hmm cant duplicate forum users issue with hm2-eth
[14:06:12] <memleak> http://dpaste.com/31R3EBX linuxcnc axis will not start, never saw this error before. branch 2.6 btw
[14:06:33] <memleak> do i need tcl3d or something?
[14:06:44] <jepler> when did it last work?
[14:07:03] <memleak> this is a new build that it never ran on before.
[14:09:16] <jepler> memleak: run this interactively:
[14:09:17] <jepler> >>> import _tkinter
[14:09:20] <jepler> >>> app = _tkinter.create()
[14:09:24] <jepler> >>> app.interpaddr
[14:09:26] <jepler> >>> app.interpaddr()
[14:09:31] <jepler> what do the last two statements print?
[14:10:08] <memleak> >>> app.interpaddr
[14:10:08] <memleak> <built-in method interpaddr of tkapp object at 0xb7308bb8>
[14:10:09] <memleak> >>> app.interpaddr()
[14:10:09] <memleak> 3109232312L
[14:10:15] <memleak> tk window shows up btw
[14:11:16] <jepler> OK
[14:11:19] <jepler> I sort of know what's going on
[14:11:38] <memleak> you always do :)
[14:15:33] <jepler> just testing here before I offer you a patch..
[14:15:54] <memleak> i compiled python with some extra security paranoid patches, thats probably the underlying cause.
[14:15:55] <jepler> what OS is this?
[14:15:57] <jepler> ah
[14:16:07] <memleak> its gentoo hardened with some extra security.
[14:16:11] <jepler> apparently nobody else's system is putting a Tcl_Interp * at an address with the top bit set
[14:18:52] <jepler> memleak: please test with http://emergent.unpythonic.net/files/sandbox/0001-axis-get-interpreter-address-the-right-way.patch
[14:20:23] <memleak> heh links was still my default browser.. will do, thanks!
[14:23:25] <memleak> works!
[14:23:30] <memleak> thanks jepler!
[14:30:20] <jepler> > Onboard LEDS show the IP address through Morse code
[14:30:25] <jepler> ooh there's an important feature
[14:30:34] <jepler> memleak: OK, will ping seb to get that included in 2.6.
[14:30:42] <jepler> seb_kuzminsky: ^^^ OK for 2.6? http://emergent.unpythonic.net/files/sandbox/0001-axis-get-interpreter-address-the-right-way.patch
[14:30:58] <memleak> glad i could help find a bug!
[14:35:15] <jepler> memleak: I'm glad it turned out to be obvious what was going on
[14:42:52] <seb_kuzminsky> jepler: yes, that looks good for 2.6. And i'm glad it was obvious to you what it was :-)
[14:45:11] <KGB-linuxcnc> 03Jeff Epler 052.6 cb5c632 06linuxcnc 10src/emc/usr_intf/axis/extensions/_toglmodule.c axis: get interpreter address the right way * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=cb5c632
[14:47:29] <seb_kuzminsky> thx
[14:47:46] <jepler> welcome
[14:48:20] <jepler> oh dear, now people are offering me ARM hardware in private e-mail
[14:49:39] <seb_kuzminsky> yay?
[14:50:14] <CaptHindsight> get my ARM board to work
[14:50:15] <jepler> seb_kuzminsky: is there a preempt-rt kernel for the odroid u3?
[14:50:29] <jepler> .. that's the one you said you have, right?
[14:51:58] <CaptHindsight> does backlash compensation work when you are using servos in closed loop? does it just get ignored or?
[14:52:19] <jepler> CaptHindsight: backlash compensation is unrelated to whether there's "real" position feedback
[14:52:34] <jepler> because the backlash could exist after the place in the system where the feedback is taken
[14:52:56] <CaptHindsight> like encoders on the servo with play in couplings
[14:52:56] <jepler> (e.g., with rotary encoders on the motor shaft, the feedback position is clearly not measuring backlash; with linear scales on the ways, probably it is)
[14:53:24] <jepler> CaptHindsight: so backlash comp is applied to commanded position regardless of feedback
[14:54:20] <jepler> down the thread, they were getting <25us latency in cyclictest with preempt-rt http://com.odroid.com/sigong/prev_forum/t1605-exploring-364-kernel-and-the-real-time-rt-patch-on-odroid-x.html
[14:54:41] <jepler> on odroid-x, not u3,
[14:55:10] <jepler> but hopefully it'd be broadly similar
[14:56:23] <seb_kuzminsky> jepler: i dont know of one, unfortunately
[15:21:56] <CaptHindsight> figures, it will probably run best on the ARM boards that are the most difficult to get
[15:27:19] <jepler> yeah, if somebody could tell me which ARM board under $200 also has a good preempt-rt kernel already packaged for debian, I'd probably click "buy it"
[15:27:36] <jepler> but as it is, I don't get any thrill out of buying a $75-$200 board and then trying to build a kernel for it
[15:27:45] <jepler> oh, and it had better build linuxcnc in more like 6 minutes than 60...
[15:28:14] <cradek> jepler: let me know if you need another real computer for $0
[15:28:19] <jepler> cradek: bah
[15:29:35] <jepler> cradek: a lot of these computers have zero fans, occupy a volume of way under a liter, and take under 10W when idle.
[15:29:41] <jepler> if you're offering that, I'll reconsider
[15:30:16] <cradek> and how long can you power one of these for $75-$200?
[15:31:17] <cradek> probably years, but not decades (or months)
[15:31:54] <cradek> that reminds me, I bought a kill-a-watt but have never used it
[15:32:23] <jepler> $200 buys you 2000kWh at $.1/kWh. 2000kWh is enough to run a 150W load for a year.
[15:32:28] <jepler> er, 1.5 years
[15:33:23] <cradek> so yeah, years but not decades. wonder what an idle pc consumes...
[15:33:40] <jepler> I think the last time I measured one of my PCs idle, it was more like 75W
[15:33:53] <jepler> take that with a grain of salt
[15:34:03] <cradek> I think our electricity is approximately .065 (winter) to .095 (summer)
[15:34:22] <cradek> I know this because I was just looking at my bill
[15:34:38] <seb_kuzminsky> the new u3 ("version 0.5") has a couple of spi ports on headers
[15:34:41] <CaptHindsight> backing up the finally working again Debian Linuxcnc file system for the cubie2 right now
[15:34:52] <jepler> yes. $0.0995 summer. $0.730 (first 30kWh/day), $0.0573 above 30kWh
[15:35:38] <cradek> so you break even in maybe 5-7 years by buying a low-power computer
[15:35:40] <jepler> seb_kuzminsky: yeah, it has spi now. voltage standard is 1.8v, so you'd need a level translator to talk to the mesa spi cards as far as I could tell
[15:35:47] <jepler> cradek: vs a free PC?
[15:35:51] <cradek> yeah
[15:36:05] <jepler> longer if you practice turning it off when not in use
[15:36:15] <cradek> not considering all the time you spend waiting for it, or shopping for faster SD cards
[15:36:36] <cradek> I can't tell if I'm being snarky. sorry if I am.
[15:36:41] <CaptHindsight> I just need it to fit into a touch screen and not need a ATX power supply
[15:37:18] <CaptHindsight> I wish the mini-itx x86 <20W boards just had +12V in
[15:37:36] <cradek> yeah power supplies are a pain.
[15:38:12] <PCW> Yeah instead of 24 pin ATX +4 pin, yuck
[15:38:14] <jepler> CaptHindsight: have you tried these kind of terrible boards? http://www.ebay.com/itm/200W-DC-12V-Pico-ATX-switch-PSU-Car-Auto-24pin-MINI-ITX-ATX-Power-Supply-FOR-PC-/290939756220?pt=PCA_UPS&hash=item43bd5c5ebc
[15:38:26] <CaptHindsight> 25uS on the Exynos is great, xenomai on the A10 was ~40uS
[15:38:42] <cradek> jepler: I have used one of those and it worked
[15:38:58] <jepler> there seem to be a zillion variations, some of them are probably totally as specified and I'm sure others are bad
[15:39:27] <cradek> car power systems are very hostile
[15:39:27] <jepler> you can even run some peripherals on this one http://www.ebay.com/itm/200W-24pin-mini-ITX-DC-ATX-car-power-supply-w-16-24V-DC-wide-range-input-F-pc/301047784739
[15:39:39] <CaptHindsight> jepler: yeah, and not bad for $28
[15:39:41] <jepler> including the standard 3.5" floppy connector, harr
[15:40:03] <PCW> I have a tiny miniITX case at home that uses somthing like that plus an external laptop wall wart
[15:42:45] <CaptHindsight> http://www.ebay.com/itm/New-DC-12V-Pico-ATX-switch-PSU-Car-Auto-MINI-ITX-ATX-Power-Supply-160W-24pin-dth-/310931121059?pt=PCA_UPS&hash=item4864f063a3 $17
[15:45:01] <CaptHindsight> http://www.pcengines.ch/ too old now, but these were all 12V
[15:45:16] <Tom_itx> http://www.mini-box.com/s.nl/it.A/id.417/.f?sc=8&category=13
[15:45:18] <CaptHindsight> used the old Geode x86 SOC's
[15:45:23] <Tom_itx> been using one of those for quite a few years now
[15:45:30] <Tom_itx> on 24/7
[15:46:05] <Tom_itx> they have others
[15:46:11] <CaptHindsight> same here $40
[15:47:14] <Tom_itx> plugged into one of these which is overkill: http://www.mini-box.com/110w-12v-8-5A-AC-DC-Power-Adapter
[15:48:27] <jepler> I think the seller doesn't have any idea what they're selling here. http://www.ebay.com/itm/-/120817594867
[15:48:34] <jepler> >
[15:48:34] <jepler> Details about New SATA ATA IDE Adapter Connector IDC 50 M/F Terminator Active Male to Female
[15:49:08] <fenugrec> I was wondering, is ClassicLadder developped outside the linuxcnc codebase then merged in ? Suppose I have patch to submit, where should it go ? (not a bug, just a minor addition I'm working on)
[15:49:22] <cradek> jepler: I don't know what that is either...
[15:49:29] <Tom_itx> hdd adapter
[15:49:33] <Tom_itx> for 2.5"
[15:49:48] <cradek> oh ok
[15:49:50] <jepler> that'd be 40 / 44 pins, not 50 pins
[15:49:59] <Tom_itx> i've got a bunch of em
[15:50:04] <jepler> saying 50 pins and "active terminator" says SCSI to me
[15:50:10] <Tom_itx> some come with a separate power connector
[15:50:15] <jepler> there have never been IDE terminators
[15:50:17] <cradek> ACTIVE on it also makes me think scsi terminator, but looks like it's male to female
[15:50:29] <cradek> I'm too tired to count the pins in the photo
[15:50:50] <jepler> 50
[15:51:08] <cradek> then I'm back to not knowing what it is
[15:51:13] <Tom_itx> you need a hdd adapter?
[15:51:29] <jepler> no, I'm actually looking for HD-50 cables for mesa cards
[15:51:51] <Tom_itx> just standard SCSI cables?
[15:51:57] <Tom_itx> flat ribbon?
[15:52:11] <jepler> oh hey, one just got delivered to me from next door
[15:52:19] <jepler> cradek: I thought we'd long since used up the ones at the office
[15:52:29] <cradek> eh there's always more junk
[15:52:31] <Tom_itx> my local guy has the connectors and sells the ribbon bulk
[15:52:50] <cradek> it's over there on the floor
[15:53:26] <cradek> (I probably made that one in 1995)
[15:54:53] <jepler> 10 MEOW
[15:54:54] <jepler> 20 GO TO 10
[16:03:05] <jepler> that killed the conversation harder than I expected
[16:08:31] <skunkworks_> playing with the 7i80?
[16:10:03] <jepler> skunkworks_: I have two of the blinkenlights daughtercards but can only find one cable
[16:11:02] <skunkworks_> neat
[16:12:50] <skunkworks_> sorry - I never got to testing the 2 new branches...
[16:15:20] <CaptHindsight> http://www.xenomai.org/pipermail/xenomai/2013-March/027974.html [Xenomai] ARM Allwinner A13
[16:15:32] <CaptHindsight> 23uS
[16:16:18] <CaptHindsight> http://www.xenomai.org/pipermail/xenomai/2013-March/027973.html
[16:19:38] <jepler> fenugrec: It's been a long time since we incorporated upstream changes from classicladder
[16:19:57] <jepler> fenugrec: if you have enhancements you'd like to see in linuxcnc in a timely fashion, better to submit them to us
[16:20:16] <fenugrec> jepler: ok. Is a "feature request" on sourceforge OK ?
[16:21:36] <CaptHindsight> http://www.xenomai.org/pipermail/xenomai/2014-March/030484.html interesting thread about Linuxcnc and xenomai on the imx.6
[16:22:00] <jepler> fenugrec: sure.
[16:24:01] <jepler> personally, I've never had to use classicladder so I'm not a good person to evaluate changes to it
[16:25:10] <fenugrec> I would just be adding a Parity and Stopbits combo-box to the Modbus config menu. The code can already handle those settings
[16:26:10] <jepler> I see.
[16:26:24] <CaptHindsight> heh the RTAI Paolo also works on xenomai
[16:27:21] <jepler> hm, I see the modbus does permit 8E1 and 8N2 so I guess it is good to offer such an option
[16:27:56] <jepler> (for RTU; 7E1 or 7N2 for ASCII, but does anybody use ASCII?)
[16:28:39] <fenugrec> jepler: yeah, currently one needs to hand-edit their CL file to add the parity setting. Parity is arguably useless since Modbus uses CRC16, but my PLC happens to want 8E1...
[16:30:13] <jepler> 8E1's not the classicladder default?
[16:30:20] <fenugrec> 8N1 is
[16:30:23] <jepler> oh that's just wrong
[16:30:42] <jepler> 8E1 is the mode that is REQUIRED, other modes are optional according to my docs
[16:31:10] <fenugrec> Really !? I hadn't even bothered to check if there was standard. Thought it was all OK
[16:31:39] <fenugrec> You're talking about "official" Modbus docs ?
[16:32:03] <jepler> I'm referring to Modbus_over_serial_line_V1_02.pdf section 2.5.1 which I found from some dodgy site on the internet years ago
[16:32:09] <fenugrec> So, do you want a patch to correct CL's default to 8E1 p-) ?
[16:32:10] <jepler> it's a 2006 document, maybe it's superseded
[16:32:37] <jepler> I dunno, would that break existing configs that work, or only new configs?
[16:32:40] <Tom_itx> <jepler> yeah, if somebody could tell me which ARM board under $200 also has a good preempt-rt kernel already packaged for debian, I'd probably click "buy it"
[16:32:48] <Tom_itx> one possibility: http://www.embeddedarm.com/software/solutions-arm.php#linux-kernel
[16:33:14] <fenugrec> jepler: possibly. I'm really not sure of the implications; I figured the best way to make everyone happy (well me at least) is to make it configurable in the GUI
[16:34:23] <jepler> Tom_itx: that looks real old .. kernel 2.4? debian woody and sarge? wow.
[16:35:45] <jepler> bbl
[16:35:47] <fenugrec> BTW, I want to compliment the linuxCNC team on its docs. They're not absolutely perfect, but jeez they're a bible compared to Mach3 docs.
[16:36:56] <jepler> by "bible" you mean "really long, and at least some people say it's riddled with contradictions"?
[16:37:13] <jepler> I mean, thanks, I know that jthornton works really hard on it
[16:37:22] <jepler> and I did a lot of work on it back in the day...
[16:38:32] <fenugrec> "and requiring some faith", yeah. Hehe, "encyclopedia" would be a better word
[16:39:11] <PCW> I can now flash a 5i25 and reload the config without rebooting
[16:40:56] <fenugrec> Mach3 packages a huge 6kB "help.txt" file with their install
[17:32:50] <CaptHindsight> 3.4.9 rtai + master work ok in Precise, ~13K after plugging in USB devices
[17:37:03] <CaptHindsight> I'll swap this drive into several systems tomorrow
[17:38:10] <CaptHindsight> is everyone leaning towards Wheezy? Anyone interested in feedback for how it works with Mint (debian or ubuntu)?
[17:42:11] <Roguish> hey all. I am looking at the 'documentation for the upcoming 2.6' and specifically at the bldc info. the mesanet 7I39 is not included in the table for the 5I20, but it is in the 5i23. an oversight possibly?
[17:42:49] <Roguish> I am setting up a system based on the 5I20 and 5I39s.
[17:43:26] <Roguish> this is the HAL manual.
[17:55:11] <PCW> Unless theres a space issue or pinout issue , pretty much any daughtercard can work with any of our FPGA cards
[18:02:26] <Roguish> it's also a matter of having the correct bit file in the correct location......
[18:02:54] <PCW> Yep, for example there s 7I77 config for a 5I20
[18:04:30] <Roguish> i can't even find where the bit files are in the 'hybrid iso' install.
[18:05:12] <PCW> not /lib/firmware/hm2/5i20 ?
[18:05:34] <PCW> you may need to install the firmware package
[18:05:50] <seb_kuzminsky> i think cradek said it's on the latest image already
[18:07:25] <Roguish> ok, there is a directory /lib/firmware/hm2 but it the icon has a little red thing on it and no files show. hidden? restricted? or me?
[18:08:55] <Roguish> permission denied!!! what's with that?
[18:09:06] <seb_kuzminsky> insufficient permissions probably. All the hm2 firmware dirs are 0700 on my machine
[18:09:22] <seb_kuzminsky> it's annoying and should be changed, but it shouldn't prevent the firmwares from loading
[18:09:47] <Roguish> ok, i'm not good at linux. how can i fix it to see the files? plesae
[18:10:41] <seb_kuzminsky> i assume you have sudo access? try "sudo ls -la /lib/firmware/hm2
[18:10:42] <seb_kuzminsky> "
[18:11:57] <Roguish> yeah that works in a terminal. how can i change the folder permissions, recursively?
[18:13:04] <Roguish> any clue what root password is on cradek's iso?
[18:13:21] <seb_kuzminsky> sudo chmod -R g+rX,o+rX /lib/firmware/hm2
[18:13:37] <seb_kuzminsky> there probably isnt one, sudo is prefered over root logins
[18:14:07] <seb_kuzminsky> you can run "sudo su -" if you want to be root for a while, but you're probably better off not doing that
[18:15:18] <CaptHindsight> runs about the same in Mint Debian Mate 17K
[18:16:43] <Roguish> ok,hey, there is a 7i39 bit file!!!! hurray!!! guess it's a doc table thing.
[18:17:04] <Roguish> in the 5i20 subdirectory.
[18:17:05] <CaptHindsight> 3.4.9 rtai + master-rt using the precise repo
[18:17:45] <seb_kuzminsky> CaptHindsight: is that 3.4-9-rtai-686-pae, version 3.4.55-3linuxcnc? that's the one in the precise (and wheezy) deb archive at linuxcnc.org
[18:17:56] <fenugrec> Roguish: my unsafe technique is "sudo passwd" and set the root password to root, then I'm all set to wreck systems :-))
[18:18:34] <Roguish> fenugrec; thanks. i will try it.
[18:20:11] <Roguish> thank you guys. now I can go back to flogging the horse i started on.
[18:20:34] <fenugrec> but seb_kuzminsky is probably right, this can cause more harm than good !
[18:20:39] <CaptHindsight> seb_kuzminsky: yes
[18:23:16] <CaptHindsight> linux-image-3.4-9-rtai-686-pae rtai-modules-3.4-9-rtai-686-pae
[18:24:01] <CaptHindsight> seb_kuzminsky: and whatever is currently in deb http://buildbot.linuxcnc.org precise master-rt
[18:24:12] <CaptHindsight> seb_kuzminsky: should I bump anything up?
[18:24:13] <seb_kuzminsky> great!
[18:24:21] <seb_kuzminsky> no, that's all the latest & greatest
[18:26:25] <CaptHindsight> tomorrow I'll try the drive in other machines
[18:34:46] <jepler> sweet, I'd never tried wake-on-lan before, but it seems to work
[18:35:33] <jepler> now I don't have to leave a certain machine on all the time just in case I want to spend 10 minutes developing linuxcnc during the day
[18:36:41] <jepler> (assuming I am happy with it booted with the default kernel, which is -rt)
[18:38:49] <fenugrec> jepler: I've been wanting to try WOL too; how much trouble was it to set up ?
[18:39:00] <CaptHindsight> if the latency doesn't get better on the A20 I might have to find a backdoor to the Exynos parts
[18:40:56] <CaptHindsight> 4412's I can get
[18:54:57] <CaptHindsight> Exynos4412 COM board $30
[19:05:11] <seb_kuzminsky> cradek: could you add kgb to the hostmot2-firmware repo?
[19:15:54] <jepler> fenugrec: not much. The computer turned out to be already enabled in BIOS. I just needed the client (package wakeonlan in debian), the HW address, and the ethernet broadcast address of my local network
[19:16:06] <jepler> wakeonlan -i 00:1c:c0:b1:5e:16
[19:16:28] <jepler> so ^^ that's how I wake that machine, your LAN broadcast address and MAC address will of course vary
[19:16:40] <jepler> about as soon as I hit enter on that command, power lights and fans start
[19:18:43] <jepler> wakeonlan is a simple program that just sends a broadcast UDP packet with magic contents: "The magic packet is a broadcast frame containing anywhere within its payload 6 bytes of all 255 (FF FF FF FF FF FF in hexadecimal), followed by sixteen repetitions of the target computer's 48-bit MAC address, for a total of 102 bytes."
[19:18:47] <andypugh> I am too lazy, and use: http://software.doogul.com/wom/ (though I doubt that would ever work for jepler)
[19:21:35] <fenugrec> jepler: nice, thanks
[19:22:42] <jepler> andypugh: you seem a bright guy, I'm not sure why you're so keen to *click* things
[19:24:13] <andypugh> Well, that bit if software saves me having to remember MAC addresses.
[19:25:17] <jepler> oh, sure, that's why I tossed the line above into a script memorably named "wake-rat"
[19:25:33] <jepler> why click when you can type "wake-rat"
[19:27:23] <jepler> andypugh: but mostly I'm just giving you a hard time
[19:28:15] <andypugh> Your attempts at “giving a hard time” are pitiful :-)
[19:47:52] <jepler> oh well
[20:18:32] <jepler> pcw_home: how many packets per servo-period should I see?
[20:18:50] <jepler> > 81c2000e0000005a
[20:18:51] <jepler> > 834200108142000d
[20:18:51] <jepler> < fee77f00ffffff00ffffff0000000000
[20:18:51] <jepler> > 83c2001000000000000000000000000081c200400000cc04
[20:20:58] <andypugh> That’s a big number
[20:27:37] <cradek> andypugh: I'm done talking to steve b for a while, but fyi, on the new wheezy cd if you go into synaptic the buildbot sources for 2.6-rt and master-rt are already listed, you only have to turn on the checkboxes you want, and then poke "install updates"
[20:30:16] <andypugh> That’s too hard and you are a bad person for making it so hard.
[20:30:33] <andypugh> You need to actually go round to his house, and explain in person.
[20:30:35] <cradek> haha
[20:31:23] <cradek> and anything other than that is wasting time and effort?
[20:31:27] <andypugh> In other news, Aram is almost showing some signs of getting a clue, he seemed to figure out halmeter all by himself!
[20:32:00] <cradek> he mostly just makes me laugh now. I don't know what he thinks he accomplishes (for himself or anyone else) by being that way.
[20:32:40] <cradek> yes I noticed he followed a signal through some stuff, which shows a big improvement in understanding some of the concepts involved. that's really great.
[20:32:48] <andypugh> “I don’t use your software because it sucked last time I tried it, but I am going to carry on hanging out here to tell you that it sucked last time I tried it”
[20:35:49] <cradek> and then ralph stirling, "I'm doing something really intricate and I dug in the code to see how it all should work, and I'm making progress"
[20:39:22] <cradek> y'know, if you've never used a real OS with a real package manager, and you expect to go clicking through web pages and running next-next-next-next-next-finish installs for everything, and handling dependencies on your own and all that, you might really be baffled by debian UNTIL you bother to try it or read about how it works
[20:39:56] <cradek> but then of course you'll be ruined and never want to go back
[20:41:17] <KGB-linuxcnc> 03Jeff Epler 05jepler/hm2-eth-v3 1a014ff 06linuxcnc 10src/hal/drivers/mesa-hostmot2/hostmot2-lowlevel.h 10src/hal/drivers/mesa-hostmot2/hostmot2.c 10src/hal/drivers/mesa-hostmot2/tram.c hostmot2: Introduce queue_{read,write} methods * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=1a014ff
[20:41:17] <KGB-linuxcnc> 03Michael Geszkiewicz 05jepler/hm2-eth-v3 5e2699c 06linuxcnc 10src/hal/drivers/mesa-hostmot2/hostmot2.c 10src/hal/drivers/mesa-hostmot2/hostmot2.h 10src/hal/drivers/mesa-hostmot2/watchdog.c hostmot2: move watchdog I/O to the read function * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=5e2699c
[20:41:17] <KGB-linuxcnc> 03Jeff Epler 05jepler/hm2-eth-v3 938bf85 06linuxcnc 10(5 files in 3 dirs) hm2_eth: New driver for Mesa 7I80 ethernet cards * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=938bf85
[20:41:28] <Tom_itx> i still find it no better or worse than windows
[20:41:53] <Tom_itx> just different
[20:45:54] <jepler> I'm so inured to everything that's terrible in linux that I barely notice it anymore
[20:46:09] <jepler> witness my spirited defense of typing commands into terminal windows...
[20:46:18] <cradek> jepler: the backspace key usually works now.
[20:46:37] <jepler> cradek: that's true, except in putty in which case it's windows's fault
[20:48:06] <cradek> seb_kuzminsky: that's a good question. I wonder how past-me did all that.
[20:57:32] <KGB-linuxcnc> 03Sebastian Kuzminsky 05master b6061b6 06hostmot2-firmware 10scripts/get-version-from-git get version from git better for other branches * 14http://git.linuxcnc.org/?p=hostmot2-firmware.git;a=commitdiff;h=b6061b6
[20:57:32] <KGB-linuxcnc> 03Sebastian Kuzminsky 05master 1c301f4 06hostmot2-firmware 10debian/rules deb: fix permissions of firmware dirs & files * 14http://git.linuxcnc.org/?p=hostmot2-firmware.git;a=commitdiff;h=1c301f4
[20:58:16] <cradek> well, that might be right...
[20:59:38] <cradek> the url is even right
[20:59:44] <cradek> that was much less hard than I expected
[21:02:59] <andypugh> Good heavens, I was sure that the Mesa BSPI driver was in 2.5.4m but no, it remains a 2.6 thing.
[21:06:18] <Tom_itx> andypugh, how's your restoration coming along?
[21:11:10] <andypugh> 2.6 does have many nice toys. Remapping is the big one as far as I can see. I seem to be using it to solve an increasing number of issues on the forum. One nice one was someone with a bunch of Mach3 files using an M998 to raise the head for tool change it was really easy to create that G-code for him. (one line in the INI and a 4-line g-code sub)
[21:11:11] <andypugh> http://www.linuxcnc.org/index.php/english/forum/40-subroutines-and-ngcgui/28031-simulate-m998-with-a-linuxcnc-macro#48464
[21:11:38] <cradek> that's awesome
[21:12:11] <cradek> I'd've used an AXIS input filter of sed, which is a much worse solution
[21:12:53] <andypugh> Tom_itx: Have you seen the blog?
[21:13:06] <Tom_itx> not for a couple weeks i'd say
[21:13:14] <Tom_itx> didn't mark it...
[21:14:44] <andypugh> Update 5 was the 6th, update 6 was last week.
[21:14:49] <andypugh> http://bodgesoc.blogspot.co.uk/2014/07/neracar5.html
[21:15:27] <Tom_itx> i think you were working on the tank last i knew
[21:15:57] <andypugh> Update 6 has the tank finished, nothing since then has been blogged.
[21:16:09] <cradek> I can't believe you sit on the tank
[21:16:14] <cradek> what an odd machine
[21:16:29] <andypugh> You sit on the seat. But the seat is suppported by the tank.
[21:16:33] <Tom_itx> you gonna ride it to work when it's done?
[21:16:33] <cradek> well, I guess my riding lawnmower was like that...
[21:16:54] <andypugh> I actually have no idea what I will do with it when it is done :-)
[21:17:26] <cradek> I identify with that
[21:17:42] <andypugh> It’s about the doing, not the having.
[21:18:01] <cradek> btw, need a watch? typewriter? radio? cnc machine?
[21:18:22] <andypugh> I might even sell it. Perhaps I could work up to a basket-case Brough Superior.
[21:20:01] <andypugh> Apart from the lack of seat (which will need tooling up for leatherworking, and making a spring-winder for 1/4” wire) I can see it being finished bar the paint this weekend.
[21:20:14] <cradek> wow
[21:21:31] <andypugh> This was last night: https://picasaweb.google.com/lh/photo/VFl7pJP9BkGcPI-GzXPDwdMTjNZETYmyPJy0liipFm0?feat=directlink
[21:22:02] <andypugh> And it now has seat supports and half a rack (ran out of MAPP gas silver-soldering the rack)
[21:24:22] <andypugh> Crikey! I should be asleep.
[21:24:26] <cradek> does that right-angle friction clutch give changing radius for something like gears?
[21:24:41] <andypugh> Yes, it has a very early CVT
[21:24:45] <Tom_itx> https://plus.google.com/photos/108164504656404380542/albums/5988166686519072513/6038643721874755954?pid=6038643721874755954&oid=108164504656404380542
[21:24:48] <Tom_itx> what are those?
[21:24:52] <cradek> cool.
[21:25:35] <andypugh> Tom_itx: A batch of fuel filler caps, and the anti-splash screes that are stuffed with horsehair (?)
[21:25:43] <Tom_itx> ahh
[21:26:05] <andypugh> Currently away at the nickel platers.
[21:26:45] <Tom_itx> how did you calculate the sprocket geometry?
[21:26:47] <andypugh> (CNC machined caps, metal-spun anti-splash)
[21:26:53] <Tom_itx> on the batch of sprockets you did
[21:27:10] <andypugh> I bought pilot-bore sprockets and machined them. They are cheaper than metal bar!
[21:27:37] <Tom_itx> years back i made a macro for my cad cam to lay spockets out
[21:27:41] <andypugh> I think a sprocket, with teeth already on, is £2.75
[21:28:09] <Tom_itx> can't make em for that
[21:28:24] <andypugh> I could in bulk, with a hob.
[21:28:39] <andypugh> But the hob is £100
[21:28:59] <andypugh> Just not worth the time really.
[21:29:21] <skunkworks> http://electronicsam.com/images/KandT/testing/stirling/DSC_1989.JPG
[21:29:56] <cradek> pretty
[21:29:58] <andypugh> My second venture in to Ner-a-Car spares is going less well than the sprockets, which sold out: http://www.ebay.co.uk/itm/Replica-Ner-a-Car-Neracar-Footboards-pair-/111408866202?pt=UK_Motorcycle_Parts_13&hash=item19f07c639a
[21:30:43] <skunkworks> not so pretty - but usable.. I tried to use a boring head to machine the connecting rod journal.. It didn't go well.. But it will work
[21:30:59] <skunkworks> (boring a hole inside out...)
[21:31:27] <cradek> negative diameter hole
[21:31:32] <skunkworks> heh
[21:31:49] <Tom_itx> andypugh, just more spares for yours
[21:31:57] <skunkworks> either I didn't have the rake correct - or the insert just wasn't right for the job.
[21:32:30] <jepler> a meaning of "journal" I was entirely unaware of
[21:32:44] <Tom_itx> skunkworks, wasn't tool deflection?
[21:33:11] <skunkworks> no - it seemed more like build up or something.. not sure.. Didn't get too many tries to fix it..
[21:33:49] <skunkworks> did I use that journal right?
[21:33:57] <andypugh> skunkworks: https://www.youtube.com/watch?v=mRuSYQ5Npek ?
[21:34:00] <skunkworks> It seemed natural...
[21:35:04] <andypugh> I have no idea what a “journalled” file system is, I assume it has lots of very teeny bearings
[21:35:26] <Tom_itx> roller bearings?
[21:35:31] <skunkworks> andypugh: that is what I though it was doing in my mind...
[21:35:34] <Tom_itx> like alot of 2strokes use?
[21:35:34] <skunkworks> ;)
[21:35:59] <skunkworks> K&T is running great
[21:37:01] <andypugh> Interesting that much of the time the material is being cut by a little bit of itself.
[21:37:18] <Tom_itx> or compressed
[21:37:33] <Tom_itx> before it shears
[21:38:27] <Tom_itx> i've seen that before but it's still interesting to watcy
[21:38:30] <Tom_itx> h
[21:39:31] <andypugh> Night all
[21:59:06] <jepler> power off: <1W, 3VA idle: 60W 60VA compiling: 105W/105VA peak running sim/axis: 75W/75VA
[22:00:24] <jepler> running 7i43, 5i20, or 7i80 @ 1kHz: 75-80W/VA
[22:00:32] <jepler> (core2 PC)
[22:00:46] <jepler> (including power supply of all 3 hm2 cards)
[22:01:43] <jepler> looks like 7i43 and 7i80 both account for 2W; I can easily unplug their power while the system's running
[22:01:48] <jepler> er, each account for 2W
[22:01:57] <jepler> biggest win: turning off computer when not in use :-P
[22:26:39] <pcw_home> jepler: should be just 3 packets. i thought the separate wd stuff was fixed a long time ago
[23:19:50] <mozmck> I've been watching this: http://www.youtube.com/watch?v=ZehZ5bq4eu0 and I'm curious linuxcnc really needs that much major work and systems replaced to be able to make remote UIs?
[23:20:03] <mozmck> curious if